Feat: Architecture docs get topic scoping, navigable code links, and in-buffer prompts

[tninja]

The C-c a A architecture document commands always regenerated the whole repository, and the documents they produced named files and functions as plain text, so there was no way to jump from a document back to the code it described. This branch narrows those commands and makes their output navigable.

• Topic scoping and in-buffer prompts. All five commands (guardrails, DDD context, test context, C4 PlantUML, repo map) now ask for an optional topic that narrows the prompt and is folded into the output file name, so a topic document no longer overwrites the repo-wide one. When the current buffer is in ai-code-prompt-mode, the prompt is inserted there as a timestamped Org section at point instead of being sent to the AI session, so it can be reviewed and edited before it goes out.
• Navigable code links. The prompt now requires the backend to link every file, module, function, variable and test case it mentions, and the command asks where those links should point. GitHub links use .../blob/HEAD/path#L42 and open in a browser; local links are relative for files inside the repository (the ../ depth is derived from the document's own output path) and absolute for files outside it, and both target ::symbol so they survive edits. The question is skipped when the repository has no GitHub remote, and an unreachable Git falls back to local links instead of failing.
• macOS test fixes. The last commit clears 13 test failures that only show up locally: session keys are canonicalized with file-truename while the tests asserted literal /tmp and /var paths (both symlinks on macOS), case-fold-search defaults to t in batch mode so a "Error:" assertion matched lower-case AppKit noise on stderr, and window-layout-transpose (Emacs 31.1) refuses layouts holding a dedicated window, which side windows are. Test-only changes, no production code involved.

Verification: the full suite reports Ran 1571 tests, 1558 results as expected, 0 unexpected, 13 skipped, and byte-compile plus checkdoc are clean on the touched files.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 54c726d335

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".