Speedup quantization in refit for FP8 GRPO

**Is your feature request related to a problem? Please describe.**
Track the effort of accelerating the in-flight quantization in refit when vllm uses FP8 precision weights.

**Describe the solution you'd like**
Quantize before transfer and do weight transfer in FP8 format; Currently we are quantizing after transfer and transfer is in BF16 format

**Describe alternatives you've considered**
A clear and concise description of any alternative solutions or features you've considered.

**Additional context**
Add any other context or screenshots about the feature request here.