When a live gbrain session held the local PGLite database, sync wrongly said the user's valid configuration was malformed. It then skipped local code and memory work with advice that could not fix the real conflict.

Sync now identifies this PGLite connect timeout as an engine lock and tells the user to stop the holding process or retry outside the live session. Malformed configurations remain a separate error, and the message does not claim PGLite's lock was removed.

Technical details

• Classify gbrain's exit 124 / connect timed out result as engine-locked for PGLite.
• Classify the equivalent non-PGLite failure as broken-db.
• Preserve broken-config for malformed configuration.
• Add actionable /sync-gbrain guidance explaining that the holder must be stopped or sync must run outside the live Claude session.
• Document that this does not remove PGLite's upstream single-process limitation.

Verification

The identical live probe was run before and after against isolated origin/main baseline and candidate environments:

• gbrain: baseline broken-config, candidate engine-locked
• memry: baseline broken-config, candidate engine-locked
• zotero-arxiv-daily: baseline broken-config, candidate engine-locked
• plane malformed-config control: broken-config before and after
• twenty malformed-config control: broken-config before and after

Every lock probe launched a real local gbrain serve; the competing gbrain sources list --json exited 124 with connect timed out.

/pr-test-env evidence gate: PASS.

Targeted tests: 36 passed, 0 failed.

Fixes #2194