dynamic distpatch of fp8 kernels (#14245)

Signed-off-by: Jeff Daily <jeff.daily@amd.com>

tests/kernels/test_triton_scaled_mm.py

@@ -32,11 +32,8 @@ def scaled_mm_torch(a: torch.Tensor,
 
 def get_8bit_types():
     types = [torch.int8]
-    supports_fp8 = current_platform.has_device_capability(89)
-    if current_platform.is_rocm() and supports_fp8:
-        types.append(torch.float8_e4m3fnuz)
-    elif current_platform.is_cuda() and supports_fp8:
-        types.append(torch.float8_e4m3fn)
+    if current_platform.supports_fp8():
+        types.append(current_platform.fp8_dtype())
     return types