Specialize relevant `cuda::(std::)` types for `__half/bfloat16/fp8`

The CUDA extended floating point types [`__half`](https://docs.nvidia.com/cuda/cuda-math-api/struct____half.html#struct____half) and [`__nv_bfloat16`](https://docs.nvidia.com/cuda/cuda-math-api/struct____nv__bfloat16.html#struct____nv__bfloat16) and fp8 (and others) are important types for many CUDA C++ developers.

As a CUDA C++ developer, I'd like it if relevant CCCL utilities like ``, `atomic`, `complex` all worked with these types.

### Tasks
- [x] https://github.com/NVIDIA/cccl/issues/1139
- [x] #3044
- [x] #3379
- [x] #3464
- [x] Specializations for `cuda::std::numeric_limits` for FP8 types
- [ ] Specializations for ``
- [ ] Specializations of complex for fp8?
- [ ] Overloads for functions for cudart extended floating-point types