[Fix] Setup announcement shows the wrong origin and LM Studio misses models on first load

[roomote]

Opened on behalf of Elliott de Launay. View the task or mention @roomote for follow-up asks.

What problem this solves

Fixes two issues users can hit during first-time extension setup. The community handoff announcement said Zoo Code picked up where Zoo Code left off, and the LM Studio settings flow could miss downloaded models until after a save and reopen.

Why this change was made

The LM Studio refresh path now keeps using the buffered settings form so the first model load reflects the user's current base URL immediately. Preview-only requests bypass the shared provider cache, including when the field is cleared back to the default localhost fallback, and the selected-model hook stays out of the base-URL refresh path so the existing debounce in settings still owns the fetch timing.

User impact

Users now see the handoff announcement credit Roo Code correctly, and LM Studio users should see their downloaded models on the first settings pass without save-and-reopen workarounds or noisy refreshes while typing.

[roomote]

1 issue outstanding: CI is still pending on platform-unit-test (windows-latest). See task

• Unsaved LM Studio base URLs now refresh the shared provider-wide cache, so previewing another endpoint can overwrite the saved runtime cache before Save.
• Including lmStudioBaseUrl in useSelectedModel() makes LM Studio model fetches fire on every keystroke, bypassing the existing debounce in ApiOptions.
• Preserving an empty baseUrl in requestLmStudioModels("") no longer reaches the preview fetch path, so clearing the field before Save still previews the previously saved endpoint.
• CI completed on platform-unit-test (windows-latest).
• CI completed on platform-unit-test (ubuntu-latest).
• CI is still pending on platform-unit-test (windows-latest) after the review wait window.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[roomote]

Fixed in e1f0f01ba. See task This follow-up keeps the LM Studio preview behavior from the earlier fixes but simplifies the handler the reviewer flagged: it passes requestedBaseUrl directly into getLMStudioModels(...) and replaces the async IIFE with a straight if/else, which keeps the failure path clearer in stack traces. Visual proof is still not applicable here because the accessible browser surface in this task is apps/web-roo-code, while the shipped change is in extension/webview messaging. Local validation passed with pnpm --dir src test core/webview/__tests__/webviewMessageHandler.spec.ts and pnpm --dir apps/web-roo-code test. The two open review comments are resolved, and the current PR checks are running on e1f0f01ba.

[taltas]

looks good