[magnusp]

Why

JDBC savepoints were a hard unsupported no-op, and xerial's SavepointTest was rejected because of that gap. This is the same change as #179, re-issued on a branch based directly on main, because #179 was merged into a stacked base branch instead of main and so never reached main.

What changed

• Implemented setSavepoint, setSavepoint with a name, rollback, and release over the transaction stream.
• Flipped supportsSavepoints to true.
• Vendored and selected the xerial SavepointTest, and added LibsqlSavepointTest with 15 tests.

Scope

• Savepoint support and the conformance re-selection. No unrelated JDBC behavior changed.

Tradeoffs

• None.

Blast Radius

• Savepoint code path in the JDBC module, plus one new vendored conformance file. The upstream test opens two connections to one path, and the conformance driver maps a path to one sqld namespace, so it passes with no divergence.

Verification

• ./tools/export-upstream --verify: verified.
• ./mvnw -pl libsql-jdbc -am test: passed.
• The conformance lanes passed on the identical content in feat(jdbc): implement savepoints and re-select the xerial SavepointTest #179, with SavepointTest 4 of 4 under both the libSQL driver and the xerial control probe.

Fixes #173