Pin the scratch store session to UTC in live tests

[erikdarlingdata]

Fixes #4639

What changed

ScratchPostgres now builds its connection string through DarlingStoreConnection.PinSessionTimeZoneUtc, the same pin the product applies. Every scratch-store live test (97 files) now runs in a UTC session whatever the cluster default is.

Why

RollupFloorCacheLiveTests reads a chunk's range_end and passes it back as $1::timestamp. In a session behind UTC that cast moves the cutoff earlier, so drop_chunks keeps the chunk and the test fails.

Test-only, not product

Every non-test store connection (NpgsqlConnection, NpgsqlDataSource.Create, NpgsqlDataSourceBuilder) in Service, Storage and Viewer goes through PinSessionTimeZoneUtc. Two Service connections in DarlingWorker (about lines 4327 and 9872) open runtime.ConnectionString without it. Both belong to monitored-target paths, not the store, so I left them alone.

Verification (local TimescaleDB rig, cluster and OS zone America/New_York)

• RED, RollupFloorCacheLiveTests on dev: 1 failed.
• GREEN, same class with the fix: 1 passed.
• Other scratch live classes on the same rig, 0 failed: IntervalFirstExecIndexRungTests (6), RollupCoverageCeilingLiveTests (1), RefreshCeilingStalenessTests (18), TempDbCeilingStoreTests (12).
• 0 failed: DocCommentHygieneTests (77), CommentFilterAdoptionTests (4), McpPayloadContractCensusTests (68), StorageCommandTimeoutTests (19), StartupCommandTimeoutTests (15). 0 build warnings.
• Not confirmed: a wildcard run of the other *Census* classes printed no totals, so CI decides those.

CHANGELOG

None: test-only change (the scratch-store test fixture pins its session to UTC).