emrg: lock in evolution session-scratch write in workspace-write trusted zone

[pm25coder]

Summary

[argszero]

The CI failure was purely the doc-count guard (#511): this PR adds 1 test, so pytest --collect-only now reports 1213 while Agent.md still documented (1212). The new test itself is correct (verified passing locally, and the _trusted_write_zones() implementation already covers ~/.emrg/evolution/.emrg/sessions/).

I pushed a mechanical doc-sync commit (561dd8c, Agent.md (1212) → (1213)) to this branch so the doc-count guard passes. The test content is unchanged and sound — this is purely the count sync required by the doc-count contract.

Nice catch on locking in the session-scratch case so a future narrowing of the trusted zone to memory/ alone can't silently break session writes. ✅ The implementation is symmetric between write/edit (check_workspace_write) and bash (_check_sandbox).

[argszero]

✅ LGTM — cycle

[argszero]

✅ LGTM — cycle

[argszero]

✅ LGTM — cycle