Fix reshard source-DB drop using empty database name

[benben]

Bug (observed on a real cnpg→cnpg reshard)

The cleanup step of an otherwise-successful cnpg→cnpg reshard failed like this (note the double space — the database name is empty):

dropping source database  on shard-A (WITH FORCE)
DROP DATABASE on the source failed — old catalog database still exists on shard-A and must be dropped manually: drop database : ERROR: zero-length delimited identifier at or near """" (SQLSTATE 42601)

The copy and cutover were fine (cleanup is deliberately non-fatal but loud), but the old catalog database was left orphaned on the source shard.

Root cause

• The admin handler fills op.SourceEndpoint/SourceUser/SourceDatabase at create time only for external sources (controlplane/admin/reshard.go); for a cnpg source they stay "".
• recordSource (runner) reads the real source info from the duckling CR status and persists it to the op row via o.fields(...) — but never updated the in-memory o.op.
• cleanupSource passed o.op.SourceDatabase (still "") to DropDatabase → DROP DATABASE "" → SQLSTATE 42601. The o.source endpoint struct built in recordSource carried the correct database all along (the copy used it successfully).

Fix

• recordSource now mirrors the persisted source_endpoint/source_user/source_database onto the in-memory o.op, keeping every later reader (cleanup logging, the end-of-op report, the ext→cnpg rollback spec) consistent.
• cleanupSource drops o.source.Database — the copy-path source of truth — and, belt-and-suspenders, logs an ERROR and skips the drop if the name is ever empty instead of issuing a zero-length identifier.

Test

Regression assertion in TestReshardHappyPathCnpgToCnpg (controlplane/provisioner/reshard_runner_test.go): the fake copier's dropdb call must name the database recorded from the duckling status. The test op's SourceDatabase is unset exactly like production, so pre-fix it fails with:

dropdb database = "", want "mdstore_org" (the recorded source database)

and passes after the fix. go build -tags kubernetes ./... and go test -tags kubernetes ./controlplane/provisioner ./controlplane/admin -run TestReshard -count=1 are green.

Note on e2e: per the resharding contract in CLAUDE.md, the cnpg→cnpg positive path (the only path that reaches cleanupSource) is unit-only until the second mw-dev shard lands, so the regression lives in the runner unit test.

🤖 Generated with Claude Code

[github-actions]

Test Impact Plan

Deterministic summary of how this PR changes tests, CI runners, and coverage-risk signals.

Summary

Area	Added	Changed	Deleted
Test files	0	1	0
E2E/journey files	0	0	0
Workflow files	0	0	0

Signals

• Test cases: +0 / -0
• Assertions: +2 / -0
• Skips or known failures added: 0
• Workflow continue-on-error added: 0
• Workflow path filters added: 0
• Test commands removed from justfile: 0
• E2E/journey retry lines added: 0

Coverage risk: neutral or increased

No coverage-reduction warnings detected.