[Misc] Load FP8 kv-cache scaling factors from checkpoints (#4893)

The 2nd PR for #4532.

This PR supports loading FP8 kv-cache scaling factors from a FP8 checkpoint (with .kv_scale parameter).

vllm/utils.py

@@ -31,6 +31,8 @@ STR_DTYPE_TO_TORCH_DTYPE = {
     "bfloat16": torch.bfloat16,
     "float": torch.float,
     "fp8": torch.uint8,
+    "fp8_e4m3": torch.uint8,
+    "fp8_e5m2": torch.uint8,
 }