fix: stop the doctor's cwd cleanup from overwriting the handshake error

[blisspixel]

Why

Last night's main CI was red for two unrelated reasons. One was a GitHub Actions incident (The job was not acquired by Runner of type hosted, Failed to resolve action download info) that cancelled most of the matrix. The other was real:

FAILED tests/test_mcp_doctor.py::TestExceptionPath::test_real_client_session_entry_failure_preserves_spawn
E  assert 'synthetic client session entry failure' in
   "PermissionError: [WinError 32] The process cannot access the file because it is
    being used by another process: '...\recon-mcp-doctor-cwd-ppzkbrfe'"

_run_handshake spawns the server with an empty tempdir as its cwd - the v1.9.3.4 cwd-shadow defense. Windows keeps that directory locked until the child has fully exited, so tearing the transport down after a failed handshake can lose the race, and TemporaryDirectory.__exit__ then raises over the top of the handshake error. The user-visible effect is worse than a flaky test: recon mcp doctor reports a file-lock message where the actual cause of the failure belongs.

What changed

• Cleanup of the isolation cwd is best-effort (ignore_cleanup_errors=True). The directory is a boundary, not state the report depends on; a leftover empty tempdir beats a masked cause. The shadow defense itself is unchanged.
• Regression test refuses to remove that directory instead of racing a real process, so it reproduces on every platform. Verified failing without the fix, passing with it.
• Separately, both roadmaps described track 3 as "measurement not started" while the network-free half of the Phase 1 baseline was already frozen in validation/2026-08-05-quality-baseline-scorecard.md, directly contradicting the paragraph below it. Corrected to Phase 1 part-run, with the corpus- and network-bound channels and the ablation still open.

Verification

uv run python scripts/check.py - all 27 stages green.

[Copilot]

Pull request overview

This PR prevents recon mcp doctor from masking handshake failures with TemporaryDirectory cleanup errors on Windows by making the isolation-cwd cleanup best-effort, and adds a deterministic regression test for the file-lock scenario. It also updates roadmap/status documentation to reflect that the Phase 1 baseline has been partially run and recorded.

Changes:

• Use tempfile.TemporaryDirectory(..., ignore_cleanup_errors=True) in _run_handshake so cleanup failures (e.g., WinError 32 directory locks) don’t overwrite the real handshake error.
• Add a platform-independent regression test that simulates the locked isolation cwd by patching os.rmdir.
• Update roadmap and changelog documentation to reflect the Phase 1 baseline’s first dated run and corrected status wording.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
tests/test_mcp_doctor.py	Adds a deterministic regression test ensuring isolation-cwd cleanup failures cannot mask handshake/session-entry failures.
src/recon_tool/mcp_client/doctor.py	Makes the isolation tempdir cleanup best-effort to preserve the original handshake error, especially on Windows.
ROADMAP.md	Updates track 3 status to “part-run” and links the first dated baseline run.
docs/roadmap.md	Aligns track 3 status wording and adds a concrete reference to the frozen network-free baseline artifact.
CHANGELOG.md	Documents the Windows cleanup-masking fix and references the new regression coverage.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.