[Bug]: LLM auto-select doesn't override unusable provider from localStorage

[wyuc]

Description

When fetchServerProviders() returns server-configured providers, the LLM auto-select logic only triggers when state.modelId is empty (line 887). But users with stale localStorage data may have a non-empty modelId pointing to a provider with no API key (neither client-provided nor server-configured).

How It Happens

1. User previously selected a model like openai:gpt-4o (stored in localStorage)
2. The v0→v1 migration only clears modelId when it's exactly gpt-4o-mini (line 952-955), missing other model IDs
3. fetchServerProviders() runs but skips auto-select because modelId is not empty
4. Generation APIs use the stale openai:gpt-4o → server has no OpenAI key → "API key required for provider: openai"

Root Cause

lib/store/settings.ts line 887 — auto-select condition !state.modelId is too narrow. It should also trigger when the current provider is not usable (no client key AND not server-configured).

Suggested Fix

In fetchServerProviders(), also auto-select when the current providerId has requiresApiKey but has no client key and is not isServerConfigured.

[nkmohit]

hi @wyuc, fix is in #126, widened auto-select to also trigger when the current provider is unusable, using the same usability rule already in the repo. verified locally with the stale localStorage repro.