fix: repair Codex local MCP install

[numnx]

Summary

Fixes the local MCP install path for Codex so the generated config points at a live Code UX HTTP gateway and can be safely reinstalled after URL or token changes.

What changed

• Only expose local MCP setup details after the Streamable HTTP gateway has actually bound.
• Replace the managed [mcp_servers.code-ux] Codex block on reinstall instead of preserving stale URL/token values.
• Clarify in the Settings MCP panel and docs that local CLI HTTP setup requires Code UX to stay running.
• Include the 0.9.3 package version bump.
• Add regression coverage for Codex reinstall behavior while preserving unrelated TOML and MCP server tables.

Root cause

The installer wrote a remote HTTP Codex entry, but setup could advertise a configured URL before the HTTP listener was live. Reinstalling also left an existing managed Codex MCP block untouched, so stale URLs or bearer tokens could survive even after the dashboard reported a successful install.

Validation

• pnpm run lint
• pnpm run test:backend
• pnpm run test:dashboard
• pnpm run build
• pnpm run check:docs-web
• Local MCP smoke: /health returned UP and authenticated initialize returned HTTP 200 with an MCP session id.

Risk / rollback

Low risk. The behavior change is limited to local MCP setup reporting and Codex TOML merge semantics for managed MCP blocks. Roll back the commit to restore append-only Codex MCP config merging.