Add support for F16 in linalg::transpose

[enp1s0]

Required by rapidsai/cuvs#716

Currently, F16 support requires users to manually insert the following code, which may not be considered user-friendly.

 namespace std { 
 template <> 
 struct is_floating_point<half> : std::true_type {}; 
 }  // namespace std 

In this PR, we define raft::is_floating_point_v<T> = std::is_floating_point_v<T> || std::is_same_v<T, half> and use it instead of std::is_floating_point_v.

Alternative 1: Define raft::is_floating_point_v<T> = std::is_floating_point_v<T> || std::is_same_v<T, half> and use it instead of std::is_floating_point_v.

Alternative 2: Add the is_floating_point<half> specialization above to some header file.

• Updade .cuh
• Update test

[enp1s0]

The current transpose passes the test for F16 due to this definition in the test code.

raft/cpp/tests/linalg/transpose.cu

Lines 34 to 37 in 98b6fe0

	namespace std {
	template <>
	struct is_floating_point<half> : std::true_type {};
	} // namespace std

[cjnolet]

Most of the primitives in raft only support 32-bit types, and I think it's really great we're starting to add support for half precision. Though I think it'll be a good idea to let the user know (in the doxygen docs) which primitives support half AND full-precision. Can you add a quick note to the doc for this transpose() function ?

[enp1s0]

Thank you @cjnolet for reviewing the code. I added a quick note.

[lowener]

I'm in favor of alternative 1, as it will lead to less confusion if we have our own definition

[enp1s0]

Thank you for your comment, @lowener. I added raft::is_floating_point. I'm not sure where to put type_traits.hpp, which defines it. Is it okay in raft/core/?