[default values] Implement ReadBridge.apply without an Iceberg fork

[cbb330]

Summary

Depends on #674 (policy), stacked on #668 → #645.

Finish the client half of the column-default read bridge: turn server-stamped config into Iceberg NestedField.initialDefault at metadata load — without forking Iceberg.

Step	Where	What
Decode	ReadBridge.from(config) in doRefresh / before IO	Parse openhouse.read-bridge.column-default.<fieldId> (kept from #668, off the retry path)
Apply	ReadBridge.apply(raw) in loadMetadata	Overlay defaults onto every schema-id, then rebuild metadata

How apply works

1. For each schema in raw.schemas(), rebuild fields with NestedField.builder + withInitialDefault when a stamped field-id is present and binds.
2. Replace the schema set on TableMetadata via public TableMetadataParser.toJson / fromJson (patch schemas / schema-id; format-v1 also patches schema). No withSchemaOverlay, no Iceberg fork.
3. Return the rebuilt metadata to the catalog load path.

Failure policy (mirrors the server encoder):

Case	Behavior
No read-bridge keys / inert bridge	Return raw unchanged
Field-id absent from a schema	Capability gap → leave that field NULL
Default cannot bind to column type	Fail loud (IllegalStateException)
Malformed known config entry	Fail loud (decode; already unrecoverable for retries via #668)
Unknown openhouse.read-bridge.* key	Ignore (forward compatible)

Semantics notes:

• Overlay hits all schema-ids (time-travel / tags / branches), not only current.
• Decode keeps validated JSON strings (shade-safe bind at apply time).
• Current stamp wins — no default history; missing ids on older schemas are gaps, not errors.

Changes

• New Features — ReadBridge.apply actually overlays initialDefault
• Internal API Changes — replaceSchemas JSON rebuild helper (public Iceberg APIs only)
• Tests — ReadBridgeTest: single/multi-schema overlay, nested fields, gaps, fail-loud bind, inert path

Why not fork Iceberg

LinkedIn Iceberg already exposes NestedField.initialDefault / withInitialDefault. The missing piece is replacing schemas on an existing TableMetadata instance. A JSON round-trip through TableMetadataParser is enough for the client overlay and keeps OH off an Iceberg fork. If Iceberg later ships a real overlay helper, swap replaceSchemas only.

Testing Done

• Added new tests for the changes made.
• Updated existing tests to reflect the changes made.
• Local: ./gradlew :integrations:java:iceberg-1.2:openhouse-java-itest:test --tests '*ReadBridgeTest'

Iceberg 1.5 / Spark 3.5 java runtime shares these sources via Gradle sourceSets (same apply path).

Additional Information

• Large PR broken into smaller PRs, and PR plan linked in the description.

Stack

1. #645 — substrate (seam + encode + decode hook)
2. #668 — decode off Iceberg’s retry path
3. #674 — policy / ramp in OpenHouse
4. #678 — sanitize-on-write (do not merge apply without this)
5. This PR — ReadBridge.apply

Depends on #678. Overlay + sanitize round-trip is covered by sanitizeAfterApplyRestoresOnDiskSchema.

Rollback Plan

Revert this PR. Clients keep decoding config (#668) but stop overlaying initialDefault; readers behave as today (NULL for missing columns). Server stamp (#645/#674) can remain inert for old clients.

[cbb330]

Restacking onto #678 (sanitize-on-write). GitHub would not change this PR's base because it is part of a stack, so apply moves to a new PR with the correct base.