feat: add native Gemini passthrough to LiteLLM sidecar

[KillerQueen-Z]

What changed

• add native Gemini generateContent and streamGenerateContent routes to the BlockRun LiteLLM sidecar
• preserve Gemini request/response JSON, native error envelopes, and SSE frames while the sidecar handles Base or Solana x402 payment
• derive streaming from the Gemini URL method instead of an OpenAI-style JSON stream field
• strip client Google credentials (x-goog-api-key, Authorization, and query parameters) before forwarding
• record the native Gemini model and /v1beta/models/... path in local usage telemetry
• document raw REST and official google-genai client setup
• bump the package version to 0.8.0

Why

The Solana gateway already supports Gemini's native protocol, but blockrun-litellm only exposed OpenAI-shaped chat and native Anthropic HTTP routes. Customers using the sidecar therefore could not point a Gemini-native client at it without translating the request themselves.

Customer impact

After upgrading blockrun-litellm, customers can point the Google Gen AI SDK custom base URL at the sidecar root and keep their existing native Gemini request/response shapes. The in-process litellm.completion(...) provider remains OpenAI-compatible and is unchanged.

Validation

• full suite: 299 passed, 2 skipped
• raw local sidecar -> Solana production E2E:
  • non-streaming native JSON returned HTTP 200 and native-ok
  • streaming native SSE returned HTTP 200 and stream-ok
• official google-genai 2.13.0 -> local sidecar -> Solana production E2E:
  • non-streaming returned sdk-native-ok
  • streaming returned sdk-stream-ok
• verified the local Solana wallet signed and settled all paid requests
• git diff --check and Python bytecode compilation pass