Support of FP8 chunk prefill

[adityachatter]

• Adds functional support of FP8 Chunk Prefill kernel
• Supports FP8 E4M3FN and E5M2 datatype. Expects Q, K, V to be in FP8 precision and descale factors for Q, K, V to be in FP32 precision with shape (batch size, number of KV heads)

Run FP8 Chunk Prefill unit tests:

cd sgl-kernel-xpu/tests
python3 -m pytest -v -s test_flash_attention.py -k dtype1

96 passed, 182 skipped, 278 deselected