Plan review menus incompatible with strict OpenAI-compatible backends — add compatibility fallback

[nguyenhoangduc0707-lang]

Problem
Plan review menus rely on model responses that include structured function/tool calls or model-specific output formats. On strict OpenAI-compatible backends (that do not expose function_call metadata or use a stricter chat-completions format), the CLI shows an empty or unusable plan review menu.

Reproduction

1. Configure CLI to use a strict OpenAI-compatible backend.
2. Run /plan and attempt to open the plan review/approval menu.
3. Menu options are missing or malformed; recommended actions are not presented.

Proposed fix

1. Add capability detection: when the LLM response lacks function_call/tool metadata, fall back to a plaintext-parsing path.
2. Fallback parsing should accept simple formats the model can produce on strict backends:
   • JSON arrays/objects if present in text
   • Numbered or bulleted lists ("1.", "- Option: ...")
   • A simple marker like ---MENU--- block or YAML front-matter
3. Prefer a deterministic parser (attempt JSON first, then list heuristics) to build the menu items.
4. Add unit and integration tests that simulate strict-backend responses (no function_call, no tool metadata) to prevent regressions.
5. Update the plan-mode e2e tests and changelog.

Why
This keeps the richer function-calling behavior when available while ensuring users on strict OpenAI-compatible backends still get usable plan review menus.

If maintainers are ok with this approach I can open a PR implementing the fallback parser and tests. Please advise if you'd prefer a stricter parsing format (e.g., requiring a special MENU: block).

[examon]

This should be resolved as of v1.0.63, whose changelog notes that plan review menus now work on strict OpenAI-compatible backends. The approval menu renders correctly on those providers again (the closely related #3716 tracked the same failure and is now closed). If you still see missing or malformed menu options after upgrading to the latest release, please comment with your provider and CLI version and we'll take another look.