Replace custom cuda hashtable with cuCollections'

## 🚀 Feature
Currently, the most expensive part of GPU based sampling is running to_block(), and specifically hash table insertions. The current implementation [cuda_hashtable.cuh](https://github.com/dmlc/dgl/blob/master/src/runtime/cuda/cuda_hashtable.cuh) does not make good use of the hardware, and instead we should replace it the implementation in [cuCollections](https://github.com/NVIDIA/cuCollections/blob/dev/include/cuco/static_map.cuh#L44).

See https://developer.nvidia.com/blog/maximizing-performance-with-massively-parallel-hash-maps-on-gpus/ for a more in-depth explanation.