feat(providers): add first-class LiteLLM Gateway preset

[Lxr-max]

Summary

LiteLLM gateways already worked as a generic custom OpenAI-compatible provider, but setup was too manual: users had to pick a wire, decide how to store a proxy key (or fail because empty keys were rejected), and read generic connection errors that did not mention LiteLLM. This PR adds a first-class LiteLLM Gateway preset in Settings so an externally hosted LiteLLM proxy can be added like other first-class gateway presets (for example CLIProxyAPI).

LiteLLM is not bundled. The app still talks to a user-hosted gateway over HTTP. BYOK and local-first storage are unchanged: proxy keys stay in local encrypted secrets, and keyless / IP-allowlist deployments store no secret.

How users add it

Settings → Models → Add Provider → LiteLLM Gateway. The modal prefills:

• Default base URL http://localhost:4000/v1
• Default wire openai-chat (user can change it)
• LiteLLM-specific help copy for master/virtual keys vs keyless allowlists
• Private-network access enabled by default for local gateways

Key paths

• Proxy-key: paste LITELLM_MASTER_KEY or a virtual key. The secret is stored encrypted. The provider entry is still marked requiresApiKey: false so it can later switch to keyless.
• Keyless / IP-allowlist: leave the key empty. No secret is stored. Activation and hasKey behave like other keyless local providers (Ollama-style).
• Generic custom providers still require an API key. Empty keys are allowed only when the client sends requiresApiKey: false (the LiteLLM preset).

Discovery and diagnostics

• config:v1:test-endpoint can probe /models with an empty key so discovery works on keyless gateways.
• 401/403, connection refused, and /models 404 get LiteLLM-specific cause/fix strings plus a hint in the connection diagnostic panel.

CI follow-up

The first runPreview with real Chrome Vitest case was racing Puppeteer's 30s launch default on busy GitHub runners and reporting Test timed out. This PR now uses a 60s Chrome launch timeout, warms Chrome once, and gives the Chrome suite a 90s budget so Lint/typecheck/test can finish. Coverage is unchanged.

Type of change

• New feature

Linked issue

Fixes #208

Checklist

• I checked the linked issue / relevant context before starting
• pnpm lint && pnpm typecheck && pnpm test passes locally
• Added/updated tests for the change
• Added a changeset (pnpm changeset) if user-visible
• Updated docs if behavior changed

Dependency additions (if any)

None. No LiteLLM runtime, binary, or SDK is added.

Principles 5b

• Compatible: existing custom-provider and CLIProxyAPI flows are unchanged; generic empty API keys are still rejected.
• Upgradeable: new preset is additive (schemaVersion unchanged); keyless flag is optional on the add-provider payload.
• Lean: reuses the Add Custom Provider modal and existing proxy-preset / diagnostics patterns instead of a new provider SDK.
• Elegant: LiteLLM-specific copy and diagnostics live next to the existing gateway preset, not a parallel settings path.

Screenshots / recordings (UI changes)

Settings Add Provider menu now includes LiteLLM Gateway; the modal shows LiteLLM help copy and an optional proxy-key field. Electron UI was not recorded in this cloud environment.