Core: Fix SnapshotUtil.schemaFor to look up schema by id

[thswlsqls]

Summary

• SnapshotUtil.schemaFor(TableMetadata, ref) read the snapshot schema with metadata.schemas().get(snapshot.schemaId()).
• metadata.schemas() returns a List<Schema>, so the schema id was used as a list index.
• When schema ids no longer match list positions, this returns the wrong schema, or throws IndexOutOfBoundsException when the id is past the list size.
• Schema ids and list positions diverge after expireSnapshots().cleanExpiredMetadata(true) removes an older unreferenced schema.
• The fix looks the schema up by id via metadata.schemasById(), mirroring the sibling schemaFor(Table, long), including its checkState guard for a missing id.
• This is a latent fix. The only in-tree production caller of this overload, MergingSnapshotProducer.apply, passes a branch ref, and branch refs return the table schema earlier in the method. The buggy line is reachable only for tag refs, so this removes an internal inconsistency rather than a live failure.

Testing done

• Added TestSnapshotUtil#schemaForTagWithMetadata (wrong-schema case) and schemaForTagWithMetadataSchemaIdBeyondListSize (IndexOutOfBounds case).
• ./gradlew :iceberg-core:test --tests "org.apache.iceberg.util.TestSnapshotUtil" — 14 tests passed.
• ./gradlew :iceberg-core:spotlessCheck :iceberg-core:checkstyleMain :iceberg-core:checkstyleTest :iceberg-core:revapi — passed (revapi backward compatible).
• Regression demo: with the fix reverted, both new tests fail (wrong schema, and ArrayIndexOutOfBoundsException); restored, both pass.