PayloadDimensionLiveTests pre-filter test asserts a cluster-wide WAL delta, so parallel live classes fail it #4354

Title

PayloadDimensionLiveTests pre-filter test asserts a cluster-wide WAL delta, so parallel live classes fail it

Body

Darling/Darling.Tests/PayloadDimensionLiveTests.cs, test
PreFilter_SkipsLockingFreshRows_ButStillRefreshesStaleOnes_AndInsertsNewDigests (added by #4288,
commit dfb4fbd, Sep 25), fails intermittently in CI's "Darling PG tests" shard 2 on PRs whose
diffs don't touch it:

PayloadDimensionLiveTests.cs:831: expected a near-zero WAL delta for an all-fresh batch, saw 39592 bytes

Seen on:

• PR Web reads stop their store query when the request is abandoned (first slice) (#4203) #4347, run 36186310095, attempt 2
• PR Bucket the Darling viewer's TempDB file I/O trend (#4234) #4353, run 36198090347

It blocks unrelated, otherwise-green PRs from merging.

Mechanism (read from code, not reproduced)

The test asserts pg_wal_lsn_diff(pg_current_wal_lsn(), before) < 1024. pg_current_wal_lsn() is
CLUSTER-wide, not scoped to the relation the test writes to. The class carries
[Collection("live-postgres")], which serializes it only against other classes in that same
xunit collection. It does not serialize against the ~39 "own-store" classes (the #1776 own-store exemption documented in LivePostgresCollectionHygieneTests.cs) that mint their own
scratch databases through ScratchPostgres and run in parallel with live-postgres classes on
the SAME Postgres cluster. Those classes' CREATE DATABASE/DDL/DML, plus ordinary autovacuum and
checkpoint activity, all write WAL on the shared cluster. Any of that traffic landing inside the
test's measurement window pushes the delta past the 1 KB bar even though the test's own
transaction touched nothing.

The own-store exemption's isolation guarantee is about rows and relations, not cluster-wide state
— LivePostgresCollectionHygieneTests.cs's comment describing the exemption should say so
explicitly.