feat(ciba): RFC 9509 backchannel authentication (Poll + Ping modes)

[thegdsks]

Summary

• Implements CIBA (OpenID Client-Initiated Backchannel Authentication, RFC 9509) for the theauth-go authorization server in Poll and Ping delivery modes; Push mode deferred
• Adds POST /oauth/bc-authorize endpoint and extends POST /oauth/token with grant_type=urn:openid:params:grant-type:ciba; exposes ApproveBackchannelAuth / DenyBackchannelAuth service methods on the root TheAuth type
• Ships CIBAStorage as an optional duck-typed interface; CIBA auto-disables at startup when the configured storage does not satisfy it, keeping deployments that do not need CIBA unaffected

Changes

New files

File	Purpose
ciba.go	Root-package re-exports: AuthenticationDevice, CIBAConfig, CIBANotification, approve/deny methods
storage_ciba.go	Root-package CIBAStorage interface alias
internal/as/ciba.go	BackchannelAuthenticate, PollBackchannelToken, ApproveBackchannelRequest, DenyBackchannelRequest
internal/as/config_ciba.go	CIBAConfig, AuthenticationDevice interface, CIBANotification, defaults
internal/as/storage_ciba.go	CIBAStorage interface (5 methods)
internal/as/token_ciba.go	MintCIBATokens helper delegating to existing mintAccessAndRefresh
internal/as/handlers/handlers_ciba.go	HTTP handler for bc-authorize; error mapping for CIBA token errors
internal/as/e2e_ciba_test.go	7 E2E tests (see below)
internal/models/models_ciba.go	BackchannelRequest struct and status/grant-type constants
internal/models/errors_ciba.go	Sentinel errors: authorization_pending, slow_down, access_denied, expired_token, etc.
storage/memory/memory_ciba.go	In-memory CIBAStorage implementation
storage/postgres/postgres_ciba.go	Postgres CIBAStorage implementation using pgx/v5
storage/postgres/migrations/0016_backchannel_requests.{up,down}.sql	Schema migration

Modified files

• internal/as/config.go: added CIBA *CIBAConfig field; validation calls applyCIBADefaults
• internal/as/dcr.go: redirect_uris no longer required for CIBA-only or client-credentials clients; added GrantTypeCIBA/GrantTypeClientCredentials/GrantTypeTokenExchange to the allowed set
• internal/as/metadata.go: ASMetadataDoc and grantTypesAdvertised conditionally populate CIBA discovery fields
• internal/as/handlers/handlers.go: mounts /oauth/bc-authorize when CIBA enabled; dispatches CIBA grant in handleToken
• as.go, wiring.go, models.go, errors.go: thin wiring and re-exports
• storage/memory/memory.go: adds ciba *cibaState to in-memory store
• storagetest/helpers.go: formatting fix

E2E test coverage

1. TestCIBAPollHappyPath: full round-trip: register, bc-authorize, pending poll, approve, token poll returns access token
2. TestCIBASlowDown: rapid consecutive polls trigger slow_down (400)
3. TestCIBAAccessDenied: DenyBackchannelAuth causes next poll to return access_denied
4. TestCIBAExpiredToken: request with 2s TTL returns expired_token after expiry
5. TestCIBARequiresLoginHint: bc-authorize with no hint returns invalid_request (400)
6. TestCIBABindingMessageInNotification: binding_message is forwarded to AuthenticationDevice.Notify
7. TestCIBAStorageMissing: storage that does not implement CIBAStorage causes 404 on bc-authorize

Test plan

• gofmt -l . returns empty (verified clean)
• go vet ./... passes (verified)
• go test -race -count=3 ./... passes (verified with 300s timeout; sleep-heavy CIBA tests need ~180s per count on CI)
• No em-dashes in any .go file (verified)
• mcpresource/ is untouched (verified via git diff origin/main -- mcpresource/)

[github-actions]

Benchmark regression report

Threshold: 25%

goos: linux
goarch: amd64
pkg: github.com/glincker/theauth-go
cpu: AMD EPYC 7763 64-Core Processor                
                     │ base-bench.txt │         pr-bench.txt          │
                     │     sec/op     │   sec/op     vs base          │
RateLimitReadHeavy-4      81.07n ± 3%   81.27n ± 4%  ~ (p=0.579 n=10)

                     │ base-bench.txt │          pr-bench.txt          │
                     │      B/op      │    B/op     vs base            │
RateLimitReadHeavy-4       0.000 ± 0%   0.000 ± 0%  ~ (p=1.000 n=10) ¹
¹ all samples are equal

                     │ base-bench.txt │          pr-bench.txt          │
                     │   allocs/op    │ allocs/op   vs base            │
RateLimitReadHeavy-4       0.000 ± 0%   0.000 ± 0%  ~ (p=1.000 n=10) ¹
¹ all samples are equal

pkg: github.com/glincker/theauth-go/crypto
               │ base-bench.txt │            pr-bench.txt            │
               │     sec/op     │   sec/op     vs base               │
Argon2Hash-4        49.79m ± 2%   47.43m ± 2%  -4.74% (p=0.000 n=10)
Argon2Verify-4      49.73m ± 2%   47.88m ± 1%  -3.72% (p=0.000 n=10)
geomean             49.76m        47.65m       -4.23%

               │ base-bench.txt │            pr-bench.txt             │
               │      B/op      │     B/op      vs base               │
Argon2Hash-4       64.01Mi ± 0%   64.01Mi ± 0%       ~ (p=0.631 n=10)
Argon2Verify-4     64.01Mi ± 0%   64.01Mi ± 0%       ~ (p=0.109 n=10)
geomean            64.01Mi        64.01Mi       -0.00%

               │ base-bench.txt │           pr-bench.txt            │
               │   allocs/op    │ allocs/op   vs base               │
Argon2Hash-4         102.0 ± 0%   102.0 ± 0%       ~ (p=1.000 n=10)
Argon2Verify-4       103.0 ± 1%   103.0 ± 1%       ~ (p=1.000 n=10)
geomean              102.5        102.5       +0.00%

pkg: github.com/glincker/theauth-go/internal/bench
                               │ base-bench.txt │            pr-bench.txt             │
                               │     sec/op     │   sec/op     vs base                │
AuditRedactor-4                     1.411µ ± 2%   1.379µ ± 0%   -2.30% (p=0.000 n=10)
OAuthCallback-4                     1.119µ ± 2%   1.006µ ± 0%  -10.06% (p=0.000 n=10)
OAuthCodeFlow-4                     49.89µ ± 1%   49.02µ ± 2%   -1.75% (p=0.002 n=10)
OAuthTokenEndpointRefreshHit-4      50.89µ ± 1%   49.67µ ± 1%   -2.40% (p=0.000 n=10)
JWKSEndpoint-4                      7.469µ ± 2%   6.784µ ± 1%   -9.18% (p=0.000 n=10)
SCIMTokenAuth-4                     22.20µ ± 3%   18.95µ ± 2%  -14.61% (p=0.000 n=10)
SessionLookup-4                    11.205µ ± 4%   9.456µ ± 2%  -15.61% (p=0.000 n=10)
geomean                             9.587µ        8.806µ        -8.15%

                               │ base-bench.txt │             pr-bench.txt              │
                               │      B/op      │     B/op      vs base                 │
AuditRedactor-4                      672.0 ± 0%     672.0 ± 0%       ~ (p=1.000 n=10) ¹
OAuthCallback-4                    1.344Ki ± 0%   1.344Ki ± 0%       ~ (p=1.000 n=10) ¹
OAuthCodeFlow-4                    15.79Ki ± 0%   15.79Ki ± 0%  -0.03% (p=0.022 n=10)
OAuthTokenEndpointRefreshHit-4     15.55Ki ± 0%   15.54Ki ± 0%  -0.01% (p=0.000 n=10)
JWKSEndpoint-4                     2.963Ki ± 0%   2.963Ki ± 0%       ~ (p=1.000 n=10) ¹
SCIMTokenAuth-4                    9.765Ki ± 0%   9.765Ki ± 0%       ~ (p=0.317 n=10)
SessionLookup-4                    8.451Ki ± 0%   8.451Ki ± 0%       ~ (p=1.000 n=10)
geomean                            4.729Ki        4.729Ki       -0.01%
¹ all samples are equal

                               │ base-bench.txt │            pr-bench.txt             │
                               │   allocs/op    │ allocs/op   vs base                 │
AuditRedactor-4                      4.000 ± 0%   4.000 ± 0%       ~ (p=1.000 n=10) ¹
OAuthCallback-4                      5.000 ± 0%   5.000 ± 0%       ~ (p=1.000 n=10) ¹
OAuthCodeFlow-4                      100.0 ± 0%   100.0 ± 0%       ~ (p=1.000 n=10) ¹
OAuthTokenEndpointRefreshHit-4       100.0 ± 0%   100.0 ± 0%       ~ (p=1.000 n=10) ¹
JWKSEndpoint-4                       37.00 ± 0%   37.00 ± 0%       ~ (p=1.000 n=10) ¹
SCIMTokenAuth-4                      56.00 ± 0%   56.00 ± 0%       ~ (p=1.000 n=10) ¹
SessionLookup-4                      42.00 ± 0%   42.00 ± 0%       ~ (p=1.000 n=10) ¹
geomean                              29.04        29.04       +0.00%
¹ all samples are equal

pkg: github.com/glincker/theauth-go/internal/jwt
            │ base-bench.txt │            pr-bench.txt            │
            │     sec/op     │   sec/op     vs base               │
JWTSign-4        31.29µ ± 1%   30.98µ ± 0%  -1.00% (p=0.000 n=10)
JWTVerify-4      72.23µ ± 0%   72.12µ ± 0%       ~ (p=0.247 n=10)
geomean          47.54µ        47.27µ       -0.58%

            │ base-bench.txt │             pr-bench.txt              │
            │      B/op      │     B/op      vs base                 │
JWTSign-4       2.782Ki ± 0%   2.782Ki ± 0%       ~ (p=1.000 n=10) ¹
JWTVerify-4     3.422Ki ± 0%   3.422Ki ± 0%       ~ (p=1.000 n=10) ¹
geomean         3.086Ki        3.086Ki       +0.00%
¹ all samples are equal

            │ base-bench.txt │            pr-bench.txt             │
            │   allocs/op    │ allocs/op   vs base                 │
JWTSign-4         15.00 ± 0%   15.00 ± 0%       ~ (p=1.000 n=10) ¹
JWTVerify-4       61.00 ± 0%   61.00 ± 0%       ~ (p=1.000 n=10) ¹
geomean           30.25        30.25       +0.00%
¹ all samples are equal

Generated by the bench workflow. A + delta means the PR branch is slower.