[magnusp]

Fixes #102

Contract: Hrana 3 carries only INTEGER, FLOAT, TEXT, and BLOB, so date/time meaning is a local SQLite convention; java.sql defines the getter surface; sqlite-jdbc is the behavioral reference, not the contract.

Base: 168f69aff4cebe175699bf5b421decf6a49c3b09
Head: fb993bd06de73bef417193faa803949cd24fe8e8

What changed

getDate, getTime, and getTimestamp now switch on the Value storage class instead of parsing getString():

• INTEGER: milliseconds since the Unix epoch.
• FLOAT: Julian day.
• TEXT: SQLite date/time formats, including yyyy-MM-dd HH:mm:ss.SSS, bare dates, bare times, optional fractional digits, and an optional time-zone suffix.
• BLOB: Bad value for type <target>; NULL: null.

LocalDate, LocalTime, and LocalDateTime are built from the same storage-class result, so getObject(column, X.class) agrees with the corresponding java.sql getter. The #69 conversion table now documents these SQLite conventions instead of text parsing.

Evidence

• Red: ./mvnw -pl libsql-jdbc test -Dtest=LibsqlResultSetTest at base + new tests -> 3 failures, all new cases (/tmp/xerial-parity-i102/lane-1-regression.log).
• Green: same command at head -> 24 tests, 0 failures; with LibsqlResultSetGetObjectTest -> 43 tests, 0 failures (/tmp/xerial-parity-i102/green-unit.log).
• Lane 2 exact inputs: TEXT 2024-01-02 10:00:00 and INTEGER 1700000000000 pass (lane-2-specific.log); xerial reference probe output recorded in reference-probe.log (TEXT -> 2024-01-02; INTEGER -> 2023-11-14 in Europe/Stockholm).
• Lane 3: all valid datetime cases assert non-null across all six targets (lane-3-specific.log).
• Lane 4 full module: BUILD SUCCESS, 238 tests (lane-4-module-suite.log).
• Lane 5 conformance: BUILD SUCCESS, 123 tests, 1 pre-existing skip, no new divergence (lane-5-conformance.log).
• Lane 6 xerial control: BUILD SUCCESS, 100 tests, 1 pre-existing skip (lane-6-control.log).
• Lane 7 spotless:check: BUILD SUCCESS (lane-7-spotless.log).
• Lane 8 full reactor: BUILD SUCCESS (lane-8-reactor.log).
• Lane 9 LibsqlResultSet*: 116 tests, 0 failures (lane-9-specific.log).
• Lane 10 package: BUILD SUCCESS (lane-10-package.log).
• Perf: ./mvnw -pl libsql-jdbc test 3x interleaved. Trunk median 2.12 s, head median 2.15 s, +1.4% (within 10%).

Divergences

• INTEGER is epoch milliseconds, not SQLite unixepoch seconds. sqlite-jdbc's default date_precision is MILLISECONDS, and the issue's 1700000000000 case requires it.
• REAL Julian days convert as a UTC instant and are then rendered in the JVM default zone. xerial decomposes the Julian day into local calendar fields; SQLite Julian days are UTC-based, and no conformance test asserts REAL.
• Bare-time TEXT uses SQLite's 2000-01-01 default date, so getDate/LocalDate return that date where xerial throws. This follows SQLite's time-value rule.
• BLOB date/time targets throw Bad value instead of xerial's numeric fallback; BLOB is not a SQLite datetime storage class.
• getTime returns a time-of-day, so Time does not carry the source date as xerial's TEXT path does. JDBC defines getTime as the time component.