feat(completion): add compact async completion menu

[fcoury]

Typing completion previously waited for both local and language-server results, then showed a large popup with an automatic preview. That made the menu feel heavier than Neovim's completion flow and left trigger characters such as . filtering an obsolete completion session while typing through an open menu.

This change makes completion immediate and unobtrusive:

• show matching buffer words immediately, then merge LSP results into the existing menu without resetting the selected candidate
• render a compact, content-sized fuzzy list with highlighted matches and LSP label descriptions instead of an automatic documentation preview
• use Neovim-style controls: Ctrl-n/Ctrl-p or arrows to navigate, Tab to accept, Ctrl-e to dismiss, and Enter to insert a newline
• preserve one-Esc exit to Normal mode, ignore late responses after dismissal, and start a fresh member-completion request on trigger characters
• default automatic completion to zero debounce while keeping completion.debounce_ms configurable

How to Test

1. Open a Python file containing an existing torch.manual_seed(...) line and add a blank line below it.
2. Enter Insert mode and type to. Expect a compact completion menu to appear immediately with buffer candidates; LSP candidates should enrich the same menu without it jumping or reopening.
3. Continue typing rch.man. Expect a fresh member-completion menu containing manual_seed and related matches, with the typed characters highlighted.
4. Press Tab, then type (1337). Expect torch.manual_seed(1337); press Esc once and expect Normal mode.
5. Regression: reopen completion, press Ctrl-e, and wait for any outstanding LSP response. Expect the menu to remain dismissed. Reopen it and press Enter; expect a newline rather than accepting the candidate.

Automated coverage includes focused completion UI/lifecycle tests plus the full Rust suite and warning-as-error Clippy validation.

[fcoury]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: aa95c39555

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".