fix(ci): qualify PRE baseline and restore full test execution

[sqlrush]

Summary

• Restore current enabled/disabled builds, all registered C unit binaries, SQL regression and static analysis.
• Fix error cleanup, invalid-output handling, exhausted bootstrap proof and byte-preserving pin refusal without weakening admission or fail-closed behavior.
• Repair portable test fixtures and compile the full unit suite in parallel.
• Make the provider-timeout fixture prove it reached the intended callback before asserting the timeout result; retain all refusal and journal checks.
• Hold the TCP fixture's per-fd EAGAIN until explicit drain/close, rather than assuming real socket saturation persists across writes; all resumed bytes and other peers still use real TCP.
• Apply the maintainer-approved MVP CI scope. Preserve the entire historical nightly suite as a separate workflow, including real failures and the 31-job baseline record.
• Require exact-commit executed fast and MVP nightly evidence before packaging; PR synthetic-merge results cannot certify a release commit.

Evidence

• Local current candidate: 310/310 unit programs, 13/13 cluster SQL tests, 6 smoke files / 172 assertions PASS.
• Policy negative tests: 9/9; analyzer execution tests: 3/3; clang-format 18.1.3: 846 files, zero violations.
• Final candidate: 5b35b5713f53880b4f9f54c6f81c635e95c9286d.
• Exact-commit fast: https://github.com/sqlrush/pgrac/actions/runs/35189193141
• Exact-commit MVP nightly: https://github.com/sqlrush/pgrac/actions/runs/35189196354
• PR merge-candidate fast: https://github.com/sqlrush/pgrac/actions/runs/35189193451
• Previous PR runs35184867813 and35185804252 exposed independent timeout-stage and TCP-backpressure fixture assumptions. Nightly35186575918 then exposed the same TCP assumption in the second-frame case, prompting a common-fixture correction rather than another site workaround. Deterministic RED/GREEN covers both (16/16 and19/19). None was waived; the final runs above must all complete successfully before merge.
• Historical full-suite result (retained, not converted into MVP PASS): https://github.com/sqlrush/pgrac/actions/runs/35179423876

No C8/soak/micro/D7/PRE campaign or retained business data was recreated. This PR publishes no release or tag. The maintainer authorized this PR's human-review bypass only after every required MVP CI job passes; main protection settings remain unchanged.

Completed integration

• All final candidate checks above passed before the authorized merge.
• Merged main: f6e213574952f81ad652cdeb76191b3a534f2fe4; its tree exactly matches the qualified candidate.
• Main fast: https://github.com/sqlrush/pgrac/actions/runs/35190031008 — 5/5 SUCCESS.
• Main MVP release CI: https://github.com/sqlrush/pgrac/actions/runs/35190041565 — 2/2 SUCCESS.
• The exact-commit release verifier accepted both main runs; protection settings were read back unchanged.
• CI cleanup is complete; no further campaign, release or tag is started.