release: prepare 0.5.0 candidate and migration guidance

[codeforester]

Summary

Prepare the immutable 0.5.0 release candidate: align VERSION and package metadata, cut the dated changelog section, document the 0.4.x migration boundary, and provide the publication/evidence checklist. The changelog validator now permits the newest VERSION-matching section to precede its tag while retaining tag checks for published history.

This draft is the final car of the 0.5.0 train, after #422. The release must also incorporate the remaining existing 0.5.0 PRs before final review. Refresh the proposed date and final changelog against the integrated release commit.

Publication is pending: independent approval, final hosted checks, TestPyPI rehearsal, annotated tag on protected main, production environment approval, PyPI/GitHub publication, and matching checksums/SBOM/attestations. No tag or package was published by this PR train. Keep the release issue open until that evidence is complete.

Refs #307.

Branch maintenance

Refs #426. Targets the benchmark branch for #422 and remains the final release car in the ordered stack. Retarget and refresh after #422 is merged; preserve the order so the release candidate is reviewed against the integrated benchmark changes.

Current-head validation

At 845e5051ebff1c7bd72adf995a27d86499a4fe9c: uv lock freshness and the full aggregate (coverage, typing, style, contracts, benchmark, static and dependency security) passed locally with all declared extras. Runtime result: 661 passed, 1 warning, and 294 subtests passed in 12.86s.

Hosted checks currently report 111/111 successful checks with no failures or pending checks. Publication evidence remains separate and is intentionally not claimed by this PR.

[codeforester]

Re-verified at 845e505: all three threads addressed ✅. The OSError path in validate_changelog.py no longer makes a second subprocess call, the README stays on published 0.4.3 until v0.5.0 is tagged, and docs/migration-0.5.md now reflects verify_discovered_config, capture_incomplete and the logger routing changes. Two release-level notes: (1) the #416 sidecar regression and the #422 benchmark-cap concern (both posted today) should be resolved before this candidate is cut; (2) #400 targets main outside this stack, so if it lands with its redaction rewrite, the changelog and migration guide should mention the redaction behavior change. The threads can be resolved.

[codeforester]

Resolved the merge conflicts in #423 and pushed commit 4cd6b8d.

Resolution summary:

• retained the 0.5.0 release documentation while incorporating #422s stdlib logging baseline and ratio budget;
• kept the ci: gate concurrent retention and logging throughput #422 benchmark calibration and stress validation updates;
• incorporated logging sidecar deletion/replacement recovery;
• added the hot-path regression test from main;
• removed all conflict markers and completed the merge against origin/main (827e4ac).

Local validation:

• full pytest suite: passed (661 passed, 2 skipped);
• focused benchmark/logging/release tests: passed;
• Ruff check and format: passed;
• strict mypy: passed;
• Markdown/docs validation: passed.

PR #423 remains open and unmerged. GitHub has accepted the branch update; hosted checks are still in progress, with no completed failures reported at this time.