Preserve the complete Windows regression with fixed pytest partitions

[taipei49314]

The Checkwash0.6.0 HOST50 regression completed its222 focused tests but timed
out during the full suite, leaving no full.xml. Preserve that failed attempt
and replace the monolithic adjudication entry with one required fixed shard
parameter0..3. Four separately accounted serial jobs must use the same exact
source and reconcile the complete Windows test population before full PASS.

Each entry records a complete ordered collection, including duplicate
occurrences, and recollects it unchanged before deterministic hash assignment.
Actual pytest invocation boundaries, final outcomes and genuine JUnit must
cover the assigned occurrences exactly; skips, xfail/xpass, missing plugins,
collection drift, timeout or incomplete raw evidence refuse. The original
eight hash-pinned wheels and five focused groups are retained. The shared
regression harness and frozen W3 assets are unchanged.

An1800-second entry deadline covers setup, tests and finalization within the
existing35-minute EC job. Internal artifacts always keep partition_complete
false; raw-digest closure and independently observed wrapper exit/elapsed are
required for partition acceptance. Late final writes and early acquisition
failures preserve incomplete observations. Raw streams are saved directly;
this ordinary regression source does not establish disk/custody adoption.

Independent source review identified three defects and separately read back
their repairs: raw logstart/logfinish, late finalization refusal, and structured
parameter/lock/plugin acquisition failure. Source CI is required. Local tests
are NOT_RUN under POLICY work-machine-local. New declaration admission and
four fresh exact HOST50 receipts remain pending; this PR runs no formal W3,
re-pin, runner adoption or measurement.