add cuda benchmark ci (#15883)

Summary:
Pull Request resolved: #15883

Introduce cuda benchmark ci for monitoring cuda backend performance.

the ci will run on three situations:

1. it will run all possible models (voxtral, gemma and whisper) on every day's 1am pst;

2. it will run an arbitrary model everytime a PR got merged;

3. manually tirggered by user.

Differential Revision: D87400561

Author: Gasoonjia