fix: model catalog follows the session's backend (TUI)

[saucam]

The TUI half of the model-catalog-doesn't-refresh-on-backend-switch bug (web half: saucam/codeoid#160).

Root cause — worse than the web

The TUI's models.list frame had no provider field at all, and the catalog was fetched exactly once at connect. So it always showed the daemon's default backend (claude) for the entire session — /provider switches and focus changes never refetched. A session on codex showed claude's models permanently.

Fix

• protocol (codeoid-protocol): ClientMessage::ModelsList gains provider: Option<String> (wire-additive, skip_serializing_if None); DaemonMessage::ModelsListResult exposes provider so the client can drop a stale result.
• fetch_models() sends the focused session's backend and runs on connect, on focus change, and after a /provider switch.
• stale-result guard: a catalog whose backend no longer matches the focused session is dropped (fast-switch race). Legacy daemons omit provider → accepted (can't guard), preserving old behavior.
• /provider optimistic update: the switch stamps the new backend onto the focused session locally before refetching, so both the fetch and its guard key off the new backend without waiting for the daemon's session.info_update frame (a separate message that may arrive after the switch response). It also clears the stale catalog so the picker can't flash the old models.

A never-used backend shows an empty catalog until its first turn (models genuinely unknown until the backend spawns) — same honest behavior as the web fix.

Tests

• result matching the focused backend is applied; stale result for the old backend is dropped; legacy (no-provider) result accepted; models.list frame carries the provider (and omits it when None).
• cargo test --workspace: 269 tui + 13 protocol + 10 client green; fmt --check clean.

Depends on the daemon already keying models.list by provider (shipped in the provider-extension work) — no daemon change needed.

🤖 Generated with Claude Code

[github-actions]

🤖 Gemini code review

This PR fixes a bug in the TUI where the model catalog failed to update upon switching backends (/provider switches or session focus changes). It adds a provider field to the ModelsList client message and the ModelsListResult daemon message, implements stale-result guarding to prevent race conditions during fast switches, and includes optimistic local state updates to prevent model picker flashing. Well-structured tests have been added to verify these changes.

Findings: 🔴 0 · 🟠 0 · 🟡 0 · 🟢 0

---

_{Tokens spent · ⬆️ Input: 4,265 · ⬇️ Output: 105 · Σ Total: 7,525
Total may be higher due to thinking token counts.}

[codecov]

Codecov Report

❌ Patch coverage is 87.31343% with 17 lines in your changes missing coverage. Please review.
✅ Project coverage is 72.15%. Comparing base (5f890ce) to head (c693cc9).
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
crates/codeoid-tui/src/app.rs	87.87%	16 Missing ⚠️
crates/codeoid-protocol/src/client.rs	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #22      +/-   ##
==========================================
+ Coverage   71.82%   72.15%   +0.32%     
==========================================
  Files          31       31              
  Lines        8434     8559     +125     
==========================================
+ Hits         6058     6176     +118     
- Misses       2376     2383       +7     

Flag	Coverage Δ
tui	72.15% <87.31%> (+0.32%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
crates/codeoid-tui/src/state/mod.rs	86.22% <100.00%> (+0.02%)	⬆️
crates/codeoid-protocol/src/client.rs	37.03% <0.00%> (ø)
crates/codeoid-tui/src/app.rs	54.06% <87.87%> (+2.64%)	⬆️