Add partial specialization dtype_traits<const T>.

[hpkfft]

If a developer provides a specialization for, say, dtype_traits<__fp16> then this commit avoids his having also to provide a specialization for dtype_traits<const __fp16>.

[hpkfft]

As an aside, the C language has _Float16 defined in technical specification ISO/IEC TS 18661-3:2015.
Arm recommends that new code use the _Float16 data type instead of the __fp16 data type.
See: Arm Half-precision

C++23 has std::float16_t as well as std::bfloat16_t. See: Fixed width floating-point types

[wjakob]

This is great, thank you!

I am happy to accept further specializations, but they must be valid in c++17.