Feat/add litellm provider

[RheagalFire]

💻 Change Type

• feat
• fix
• refactor
• perf
• style
• test
• docs
• ci
• chore
• build

🔀 Description of Change

Adds LiteLLM as a first-class AI gateway provider, giving NextChat users access to 100+ LLM providers (AWS Bedrock, Vertex AI, Groq, Mistral, Cohere, etc.) through a single proxy endpoint.

Changes (8 files, 451 insertions):

New files:

• app/client/platforms/litellm.ts - Client-side provider implementing LLMApi (chat with streaming, model auto-discovery via /v1/models, tool calling, reasoning content support)
• app/api/litellm.ts - Server-side proxy handler with auth and model filtering
• test/litellm.test.ts - 13 unit tests (path resolution, message extraction, model discovery, error handling, chat payload)

Modified files:

• app/constant.ts - Added LITELLM_BASE_URL, LiteLLM to ServiceProvider/ModelProvider enums, ApiPath.LiteLLM, LiteLLMConst config
• app/client/api.ts - Imported LiteLLMApi, added dispatch cases in ClientApi and getClientApi()
• app/store/access.ts - Added litellmUrl, litellmApiKey state, isValidLiteLLM() validation
• app/config/server.ts - Added LITELLM_URL, LITELLM_API_KEY env vars
• app/api/auth.ts - Added LiteLLM auth case
• app/api/[provider]/[...path]/route.ts - Registered LiteLLM route handler

Tests*

Unit tests (13/13 pass):

PASS test/litellm.test.ts
  LiteLLMApi
    path()
      ✓ builds correct path with custom config URL
      ✓ strips trailing slash from base URL
    extractMessage()
      ✓ extracts content from standard response
      ✓ returns empty string for missing choices
      ✓ returns empty string for null content
    models()
      ✓ fetches and parses model list from /v1/models
      ✓ returns empty array on HTTP error
      ✓ returns empty array when data is null
      ✓ preserves provider/model format in model IDs
    usage()
      ✓ returns zero usage
    speech()
      ✓ throws not implemented error
    chat()
      ✓ sends correct payload for non-streaming
      ✓ calls onError when fetch fails
Tests: 13 passed, 13 total

Live E2E (Azure Foundry Claude Sonnet 4.6 via LiteLLM proxy):

--- Chat completion ---
Status: 200, Content: "10", Usage: prompt=20, completion=5, total=25
--- Model auto-discovery ---
Models found: 1 (anthropic/claude-sonnet-4-6)
--- Response structure ---
id: chatcmpl-..., object: chat.completion, role: assistant, finish_reason: stop

📝 Additional Information

• Additive only - all 14 existing providers untouched
• No new npm dependencies
• Follows the SiliconFlow provider pattern exactly
• Server-side deployment supports LITELLM_URL and LITELLM_API_KEY env vars
• Default proxy URL http://localhost:4000 matches LiteLLM's standard port
• Models auto-populate from the proxy's /v1/models endpoint

Client-side setup:

1. Select "LiteLLM" as the provider in Settings
2. Enter your LiteLLM proxy URL (default: http://localhost:4000)
3. Enter your LiteLLM API key

Server-side (env vars):

LITELLM_URL=http://your-litellm-proxy:4000
LITELLM_API_KEY=sk-your-litellm-key