Predicate: is the dtype a real floating-point type?
Dtype::C64 is not floating point under this predicate even
though its lanes are float32 — complex tensors take a separate dispatch
path in most kernels.
Parameters
dtDtypeDtype to test.
Returns
booltrue for F16, F32, F64; false otherwise.
Notes
Must be kept in sync with the Dtype enum: if BF16 is added,
it should be included here.