[FEA] Block level bulk APIs

[devavret]

Currently the bulk APIs are host calls which operate on one map. We have a use case where we have multiple small maps defined on a slice of pre-allocated slot space. For this we duplicate the insert kernel code from cuco and initializing one map per block.

It would be nice to have cub style block level bulk APIs for initialize, insert etc.

[jrhemstad]

Love this idea. For initialization we do already have the equivalent of this functionality with make_from_uninitialized_slots: https://github.com/NVIDIA/cuCollections/blob/fb58a38701f1c24ecfe07d8f1f208bbe80930da5/include/cuco/static_map.cuh#L716-L727