[clang] [CUDA][HIP] Make template implicitly host device (PR #70369)

[Artem Belevich via cfe-commits]

Artem-B wrote:

Now that we're making an even larger class of functions implicitly HD, the last logical step would be to make *all* unattributed functions implicitly HD, too (in a separate patch). After all, a template is as GPU-portable (or not) as a regular function. Unlike constexpr or compiler-generated glue for lambdas, template functions do not confer any benefits to our assumptions about  whether the code will be compileable and working on a GPU.


https://github.com/llvm/llvm-project/pull/70369