### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

When an MCP server using Streamable HTTP transport (transport: http) and OAuth2 client credentials (auth_type: oauth2) is registered via the REST API (POST /v1/mcp/server), tool discovery always fails silently. The server is created successfully (HTTP 201), background health check runs and sets last_health_check, but status stays permanently unknown and no tools are ever returned from that server. The LiteLLM pod logs show asyncio.CancelledError — the async task running the Streamable HTTP client is externally cancelled before the OAuth2 token exchange and tool listing can complete.

The same server configuration works correctly when registered via the ConfigMap mcp_servers: block — all tools are discovered at startup with no errors.

### Steps to Reproduce

1. Deploy LiteLLM Enterprise with a data plane (control plane / data plane split). 2. Register an MCP server via REST API with transport: http and auth_type: oauth2:

``` POST /v1/mcp/server { "server_name": "DatabricksSQL", "transport": "http", "url": "https://<workspace>.azuredatabr…