Spec create/update no longer hangs, returns a clear “locked” error

Creating or updating a spec via the MCP tools or the dashboard could hang indefinitely: validation passed, then the server wedged on an internal advisory lock that had no timeout, and clients timed out without a clear answer. The fix bounds the write transaction with a short lock_timeout and converts a contended write into a structured {"error":"locked"} response on the MCP create_spec / update_spec tools, plus a “being written, try again” retry flash on the spec new/edit/show LiveViews, so the operation fails fast and the user can retry instead of hanging or crashing. Deploying this release also clears any existing wedge, because the stuck Postgres connection’s locks are released on the BEAM restart.