Fix OpenAI SDK request forwarding

[Danielxu0208]

Summary

• forward the effective generation temperature to OpenAI SDK requests
• delegate OpenAISDKStreaming(stream=False) to the parent request path with the correct signature
• avoid duplicate preprocessing and rate-limit acquisition in the non-streaming fallback
• add request-level mock regression coverage

Why

OpenAISDK._generate received the effective temperature from generate but sent self.temperature, which is normally None. The streaming adapter's disabled-stream fallback also passed an extra positional timeout argument to the parent method, so generation failed with TypeError despite initialization succeeding.

Validation

• default and per-call temperatures reach the SDK request
• constructor temperature keeps precedence over per-call values
• stream=False returns a normal response and does not send the streaming flag
• relevant pre-commit hooks passed
• modified files compile successfully