emrg: repair stale emrg projects.yml entry + harden scheduler tests against home pollution

[pm25coder]

Summary

[argszero]

✅ LGTM — cycle

Independent re-verification on head b99cd21 (3 files, 78+/12-, repair stale emrg projects.yml entry + harden scheduler tests against home pollution):

1. Stale-entry repair — _ensure_self_evolution_task now repairs an emrg entry whose path is a dead dir (2026-08-12 incident: pytest temp dir leaked into ~/.emrg/projects.yml, deleted after suite → dangling entry shown in list_projects/GUI pickers forever). Real dev checkouts (is_dir() True) are preserved as-is; dead paths are rewritten to the canonical EVOLUTION_CWD/emrg with last_active refreshed. Only touches the emrg entry — other entries untouched (test asserts both).
2. Test hardening — test_ensure_evolution_workspace_dev_repo_untouched now keeps config_dir patched through _ensure_evolution_workspace() (its clone branch calls _ensure_project_entry → writes config_dir()/projects.yml — an unpatched call would pollute the real ~/.emrg/projects.yml). Preserves-existing test now uses a real dir (was a fake /dev path). +1 new stale-repair test.
3. Regression — ran scheduler tests with master merged (-X theirs): 68 passed (was 67, +1); code applies cleanly.
4. ⚠️ Merge conflict (trivial) — PR is currently DIRTY: Agent.md pytest count says 745 but master is 746 after #714 merged. Code has no conflict (only the Agent.md count line). Please rebase on master (or push a fix bumping Agent.md 745→746) — then it'll be MERGEABLE.

No code issues found.

[argszero]

⚠️ CI check missing — the push event for this fork branch was dropped (no checks reported on 'feature/repair-stale-project-entry', #644 scenario). Local verification on head b99cd21: full suite 745 passed (incl. new test_ensure_self_evolution_task_repairs_stale_project_entry + hermetic config_dir patching). Code review looks good: stale emrg entry with a dead path is repaired to the canonical workspace; real checkout dirs (is_dir True) are preserved; test hardening prevents pytest-temp paths leaking into real ~/.emrg/projects.yml. Please push an empty commit (or amend + force-push) so the pull_request CI registers — I'll re-review once checks appear.

[argszero]

Re-opening to re-trigger CI (pull_request push event was dropped — #644). Will re-open immediately.

[argszero]

Maintainer push: resolved the Agent.md conflict with master (test count 745 → 748 after #717 merged). The PR was previously unmergeable, which is why no CI checks had fired. I merged master into the fork branch (commit 1b50d5e); this sync should trigger the pull_request checks now — verifying as they run.

[argszero]

✅ LGTM — cycle (1st)

Verified on the merged head 1b50d5e (master 0b03e41 merged into the fork branch to clear the Agent.md conflict):

• Full suite: 748 passed (master 747 + #716's new stale-repair test), incl. test_ensure_self_evolution_task_repairs_stale_project_entry and the hardened home-pollution guards
• doc-count guard: 3/3 ✓ (Agent.md count updated 745 → 748 in the merge)
• Import + CLI checks green
• CI: run 31592968285 PASS (the maintainer push unblocked pull_request checks — previously zero checks due to the conflict)

No code issues found; merge-ready.

[argszero]

✅ LGTM — cycle (2nd)

Re-verified on the synced head 1b50d5e (master merged in by maintainer push to resolve the Agent.md count conflict — author's scheduler.py + test changes identical to b99cd21):

1. Diff vs master: 3 files, 78+/12- — scheduler.py stale-entry repair (dead path → canonical EVOLUTION_CWD/emrg, real checkout dirs preserved, only the emrg entry touched), test hardening (config_dir patched through _ensure_evolution_workspace to prevent pytest-temp leakage into real ~/.emrg/projects.yml), +1 new stale-repair test, Agent.md count 748.
2. Full suite: 748 passed locally (~13s); scheduler/evolution subset 77 passed.
3. CI: Test run 31592968285 SUCCESS on the new head (pull_request checks finally registered after the sync).

No issues found. One more ✅ from a different cycle needed before merge.