release: v3.14.6 — silent AP self-heal + canonical domain ids

End-user-visible improvements bundled into a single patch release so
``uvx upgrade`` + ``cortex-setup-project`` works first time on a clean
install with zero hand-holding:

* fix(domains): worktree paths and slug-tail tokens
  (voice/body/for/via/few/large) no longer leak as orphan domain hubs in
  the visualisation. Profile rebuild collapses 40 path-noise keys down
  to ~8 canonical ones; workflow-graph builder buckets non-canonical
  free text under ``domain:__global__``. (commit 81c74ff)

* fix(ap): ``ingest_codebase`` silently pre-cleans stale single-file
  ``graph`` slots before invoking AP, and retries once if AP races on
  the same slot mid-call. ``resolve_graph_paths`` now scans BOTH
  ``~/.cortex/ap_graphs/`` and ``~/.cache/cortex/code-graphs/`` so the
  visualisation finds graphs regardless of which AP version wrote them.
  (commit fa2c16c)

Either failure used to surface as a hard error to the user during
first-run setup. Both are now silent recoveries.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: cdeust

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "neuro-cortex-memory"
-version = "3.14.5"
+version = "3.14.6"
 description = "Scientifically-grounded memory system based on computational neuroscience research"
 readme = "README.md"
 license = "MIT"