fix: keep task writer transitions committable

[SUaDtL]

Summary

• Require queued tasks to enter in-progress before done, with actionable no-write errors for invalid transitions.
• Keep every successful public task mutation acceptable to the commit classifier, including ID minting and missing-section adds.
• Reject malformed namespaces and add fields before they can inject invalid board lines.
• Reject internal queued-state rewinds that could erase started or completed evidence.
• Retry only the exact transient Windows proof-admission refusal once, without changing production containment.
• Regenerate the Claude, Codex, and Pi task surfaces from the shared core.

Why

Issue #364 exposed a contract split: the sanctioned writer could report success for output that the commit gate refused. This change restores one ordered lifecycle and makes success mean the resulting board can ride its work commit.

The implementation follows ADR-0008: preserve strict lifecycle atomicity, classify only exact sanctioned writer shapes, and reject malformed input before mutation.

This PR is stacked on #365. It now closes the inherited internal queued-rewind finding in #366 and carries the proof-only hosted hardening from #339.

Test plan

• python .github/scripts/test_taskwriter.py (46 passed)
• python .github/scripts/test_taskboardlib.py (104 passed)
• python .github/scripts/test_board_sync.py (6 passed)
• Full configured Python gate, including 1,023 hook tests
• Pi tools typecheck and Vitest suite (545 passed, 1 intentional skip)
• python .github/scripts/test_pi_platform_contract.py --fixtures-only
• Core sync, surface generation, host package, reference graph, compilation, secret, and diff checks
• Independent contract review and required coverage audit, both PASS with no CRITICAL or HIGH findings

Closes #364
Closes #366
Closes #339

[SUaDtL]

Backlog reconciliation follow-up (2026-07-20):\n\n�2.rev.0026 and �2.rev.0027 are already complete in authoritative history but remain queued on the board:\n- 66eca56 / PR #201 records the scoped LGPL-3.0-or-later + 0BSD decision and CONFIRM-09 strategy.\n- 23fa77a / PR #216 implements and releases the audit staleness warning in ca 2.8.11.\n\nA fresh ADR-0008 backstop attempt reproduced the contract mismatch this PR fixes: the currently installed writer reports queued-to-done success, while main's classify_board_diff rejects that transition. A combined two-task flip is also rejected because the classifier intentionally accepts exactly one transition per work commit.\n\nPost-merge recovery: reconcile each task separately through the landed lifecycle (start, then done), one task-writer transition per evidence-bearing work commit. Do not bypass or broaden the classifier. No source, dependency, runtime, or security-posture change is required.

[SUaDtL]

Superseded by #313, the single hackathon review surface. This PR's recorded head and changes are represented in #313's exact 16-PR / 29-commit / 126-path source manifest. PR #313 passed final exact-head CI at commit 6173b1d and remains open and unmerged. Closing this source PR to avoid parallel review; please continue review and discussion on #313.