Payload-dimension pre-filter test no longer asserts a cluster-wide WAL delta (#4354)

[erikdarlingdata]

Closes #4354.

Why

PayloadDimensionLiveTests.PreFilter_SkipsLockingFreshRows_ButStillRefreshesStaleOnes_AndInsertsNewDigests
(added by #4288) asserted a near-zero pg_wal_lsn_diff(pg_current_wal_lsn(), before) for an
all-fresh batch. pg_current_wal_lsn() is cluster-wide, not scoped to the relation the test
writes. The class is [Collection("live-postgres")], which only serializes against other
live-postgres classes — it does not serialize against the ~39 "own-store" classes (the #1776 own-store exemption) that mint scratch databases on the SAME cluster and run in parallel. Any of
that traffic (plus autovacuum/checkpoints) landing inside the test's measurement window pushes the
delta past the 1 KB bar even when the test's own transaction touched nothing. Seen on PR #4347
(run 36186310095, attempt 2) and PR #4353 (run 36198090347), both unrelated to this test.

What changes

Test-only. No production code changes.

• Dropped the pg_wal_lsn_diff assertion and its two now-unused helper locals
  (CurrentWalLsnAsync, WalBytesSinceAsync) from
  PreFilter_SkipsLockingFreshRows_ButStillRefreshesStaleOnes_AndInsertsNewDigests. The test's
  existing xmax <> 0 and last_seen checks already pin "no lock taken, no row touched" directly
  against the rows this test wrote — they have no exposure to unrelated cluster WAL traffic, so
  the WAL assertion was redundant as well as flaky. The summary doc comment on the test now
  explains why the WAL check was there and why it was removed.
• LivePostgresCollectionHygieneTests.cs's doc comment on the own-store exemption now says
  explicitly that the exemption's isolation is scoped to rows/relations, not cluster-wide state
  such as WAL position, which every session on the same server moves.

Test plan

• dotnet build Darling/Darling.Tests/Darling.Tests.csproj -p:EnableWindowsTargeting=true: 0 errors, 0 warnings.
• Red/green proof, against a local timescale/timescaledb:2.30.1-pg18 container and a throwaway
  net10.0 console harness (outside the repo) that replayed the test's arrange/act steps via
  PgMigrations.MigrateAsync + PayloadDimensionWriter.FlushAsync, with concurrent WAL noise from
  another session standing in for parallel own-store classes:
  • Pre-filter ON (current shape): locked=0, changedLastSeen=0 — PASS.
  • Pre-filter mutated OFF in a local, reverted copy of PayloadDimensions.UpsertSql (dropped the
    WHERE NOT EXISTS anti-join, restoring the pre-Payload-dim upsert still writes on every sighting: ON CONFLICT DO UPDATE locks each presented row before its one-hour guard skips the UPDATE #4249 shape): locked=50, changedLastSeen=0 —
    FAIL on the xmax check, confirming the remaining assertions still catch the regression Payload-dim upsert still writes on every sighting: ON CONFLICT DO UPDATE locks each presented row before its one-hour guard skips the UPDATE #4249
    fixed without the WAL bar. The mutation was reverted; git status shows only the two files
    listed above changed.
• PayloadDimensionLiveTests and LivePostgresCollectionHygieneTests against the shipped
  runtime (PG 18.6 + TimescaleDB 2.30.1) — Windows-only (net10.0-windows), left to CI.
• Full Darling.Tests suite — Windows-only, left to CI.

For the coordinator

• The harness used a bundled TimescaleDB 2.30.1-pg18 image, matching the shipped runtime pin per
  prior lessons on this repo, but is otherwise a throwaway rig (not CI, not the fetch-pg-runtime.ps1
  path) — treat the red/green numbers as a code-level proof of the assertion's behavior, not a CI
  run.
• No other test in the touched files was changed.