feat(discord): searchable /model autocomplete

[FixItFoundry]

What

Adds type-to-filter autocomplete to the Discord /model slash command, so users can search the full set of authenticated models instead of scrolling a capped select menu.

Why

/model's picker is a Discord select menu, which Discord caps at 25 options (and the current picker surfaces only the first ~50). Once several providers are authenticated — each with dozens of models — the picker becomes impractical: no search, and most models aren't even shown.

How

• _model_autocomplete_entries() builds a flat (label, value) catalog from list_authenticated_providers() (the same source the picker uses), uncapped, with the current provider ranked first. Cached ~120s since autocomplete fires on every keystroke; the underlying read is served from the on-disk provider-model cache.
• _autocomplete_model callback filters by the typed substring and returns up to Discord's 25-choice limit. It reuses the existing slash authorization (_evaluate_slash_authorization) so the catalog isn't leaked to unauthorized users, and it never raises — any failure yields an empty suggestion list.
• @app_commands.autocomplete(name=...) wired onto the /model command.

Compatibility

Fully backward compatible. The stock select-menu picker remains as the fallback (empty query / no autocomplete client), and every failure path degrades to an empty suggestion list, so command registration and handling are unaffected.

Testing

Verified across multiple agents and pulls all models from all wired providers. Typing filters live; an empty query surfaces the current provider's models first; unauthorized users get no suggestions.

[teknium1]

Thanks for addressing a real Discord /model usability gap; current main still has a plain string-only /model command at plugins/platforms/discord/adapter.py:4082-4085 and the picker caps its selects at 25 entries (plugins/platforms/discord/adapter.py:7251-7255, 7274-7297).

Problems

• Blocking: plugins/platforms/discord/adapter.py:4010 calls the synchronous catalog helper from the async autocomplete callback. That helper calls list_authenticated_providers at PR line 3932. Current gateway /model deliberately uses await asyncio.to_thread(...) for this category of listing (gateway/slash_commands.py:1514-1526) because a stale cache can take the synchronous HTTP path; the regression test documents prior 120–150-second gateway freezes (tests/gateway/test_model_command_async_offload.py:4-19).

Suggested changes

• Offload _model_autocomplete_entries() with await asyncio.to_thread(...) before filtering the returned entries.
• Add regression coverage for decorator registration, authorization returning [], filtering/capping, and the non-blocking listing boundary. tests/gateway/test_discord_slash_auth.py:673-719 provides a decorator-capture pattern.

This is an automated hermes-sweeper review.

[FixItFoundry]

I recently changed my GitHub username from jessecasco to FixItFoundry. The workflow run for this commit was flagged/blocked by the CI security policy since the namespace changed, so it might need manual approval to run.