v0.4.2 — MCP run resources give metadata, not run output

Upgrading from 0.4.1: back up the database and OpenBao, then docker compose build and docker compose up -d. No migration and nothing to set. If up -d stops with dependency failed to start: container alexclaw-openbao-init-1 exited (0), AlexClaw is left created, not started: once migrate has exited 0 (docker compose ps -a), start it with docker compose up -d --no-deps alexclaw-prod.

Changes

• MCP run resources: alexclaw://runs/list and alexclaw://runs/{id} give a run's metadata — status, timing, node, each step's name, skill and outcome, and the kind of a failure. They no longer return a run's result, a step's output or an error's text, which could hold the data a step read (a sql_query step's rows) from any run, whatever started it. An MCP client receives output only as the answer to a run it started itself. See SECURITY.md, "Known Limitations and Design Decisions".
• Workflows page: the Enabled tooltip says a disabled workflow does not run at all, and that imported workflows arrive disabled.
• Documentation: the public docs are checked against the code in the test suite — every command, script and path a doc names exists; README's skills and admin-page tables match the code; stated counts match the source; every internal link resolves.