Feat/add litellm provider

[RheagalFire]

Summary

• Adds LiteLLM as a first-class provider type in Neo Chat, giving users access to 100+ LLM providers (AWS Bedrock, Azure, Vertex AI, Cohere, Replicate, etc.) through a single proxy endpoint.
• Includes proper base URL resolution, outbound security validation, and comprehensive tests.
• Fixes a pre-existing bug where createOpenAIClient() hardcoded "OpenAI" for base URL resolution instead of using the actual provider type.

Test Plan

• pnpm format:check
• pnpm lint
• pnpm typecheck
• pnpm test (850 passed, 171 suites)
• pnpm build

Notes

• I updated docs or examples when behavior changed.
• I considered local-first data handling and hosted deployment safety.
• I added or updated tests for user-facing or security-sensitive changes.
• For localization changes, I followed docs/localization-pr-guide.md and documented intentional English fallbacks.\

Changes

Provider registration (7 files):

• src/lib/providers/types.ts - Added "LiteLLM" to ProviderType union
• src/lib/providers/providerTypes.ts - Added LITELLM_PROVIDER_TYPE constant, updated isProviderType() and isOpenAIProviderType()
• src/lib/api/schemas.ts - Added "LiteLLM" to the Zod runtime validation enum
• src/lib/security/urlPolicy.ts - Added default LiteLLM base URL (http://localhost:4000/v1)
• src/utils/apiHelpers.ts - Added "LiteLLM" to the ProviderConfig type union
• src/components/settings/ProviderSettings.tsx - Added LiteLLM to provider type dropdown

Tests (4 files, +100 lines):

• src/__tests__/providerRuntimeSchema.test.ts - LiteLLM schema validation
• src/__tests__/urlPolicy.test.ts - 5 tests: default URL, custom URL with/without /v1, models path
• src/__tests__/providerConfig.test.ts - 1 test: normalizeModelProvider preserves LiteLLM type
• src/__tests__/providerOutbound.test.ts - 3 tests: client creation, localhost blocked in hosted mode, public URL allowed

Example usage

// In your app configuration or via the Settings UI:
const provider = {
  type: "LiteLLM" as const,
  baseUrl: "http://localhost:4000/v1",  // default, auto-filled in UI
  apiKey: "sk-your-litellm-key",        // or leave blank if no auth
};

// Models are auto-discovered from the proxy's /v1/models endpoint.
// Any model your LiteLLM proxy serves becomes available:
//   anthropic/claude-sonnet-4-6
//   bedrock/anthropic.claude-v2
//   vertex_ai/gemini-pro

Risk / Compatibility

• Additive only. No existing provider types or streaming logic modified.
• No new dependencies. LiteLLM proxy speaks the OpenAI protocol.
• Hosted deployment safe. LiteLLM's default localhost:4000 is correctly blocked in hosted mode (tested).
• Base URL fix benefits all providers. The provider.type fix is a correctness improvement for any future provider type.

[Amery2010]

Thank you for your support of this project.

The project currently supports OpenAI-compatible API formats, which are already a common format and support local models such as Ollama. Personally, I don't think it's necessary to add a general-purpose SDK, as this would make it difficult for users to choose the appropriate API type.