feat: support fix speculative rating for performance debug.

[a120092009]

No description provided.

[zhang-minchao]

/gemini review

[gemini-code-assist]

Code Review

This pull request introduces a RejectionSamplerRateController to allow fixing the speculative acceptance rate for performance debugging, controlled via an environment variable. The implementation adds a complex controller with PID logic and various heuristics. While the feature appears to be implemented as intended, the new controller class has significant code quality issues, primarily due to a large number of "magic numbers" that make the code difficult to understand, maintain, and tune. My review focuses on improving maintainability and robustness by suggesting the use of named constants and safer coding practices.

[zhang-minchao]

/gemini review

[gemini-code-assist]

Code Review

This pull request introduces a feature for performance debugging of speculative decoding by allowing a fixed acceptance rate to be set via an environment variable. This is implemented through a new RejectionSamplerRateController class that is used by the RejectionSampler. The changes are well-structured, introducing a new utility function to parse environment variables and integrating the new controller cleanly. My review focuses on improving maintainability by replacing magic numbers with their corresponding named constants that are already defined in the header file.