feat(providers): manual model ping and selective validation checks

[cat1949]

Summary

Improve provider configuration UX and safety for OpenAI-compatible chat providers:

• avoid automatic chat ping unless model is explicitly selected
• allow manual model ping from provider settings

Changes

• Add model select/manual input in provider settings page
• Add manual validation trigger (runValidationNow) and UI button (??????)
• Update OpenAI-compatible validators:
  • connectivity/model list checks use model-list request path
  • chat-completions check runs only when config.model is set

Why

Prevents accidental, uncontrolled chat ping behavior during provider setup and reduces risk of rate-limit spikes.

Testing

• pnpm -rF @proj-airi/stage-ui typecheck
• pnpm -rF @proj-airi/stage-pages typecheck

[garrytan]

Your fix was independently shipped via another route before this PR could land — verified in current main. Closing. Thanks for the contribution, @spacegeologist!

[spacegeologist]

Thanks Garry! Glad the fix made it into main. Happy to pick up another small browse/gstack issue if there is one you think would be useful.