[ENHANCEMENT]: Use `static_for` to guarantee compile-time loop unrolling

[PointKernel]

Is your feature request related to a problem? Please describe.

Currently, cuco relies on #pragma unroll to hint the compiler for more efficient code generation in fixed-size for loops, for example:

cuCollections/include/cuco/detail/open_addressing/open_addressing_ref_impl.cuh

Lines 980 to 981 in 3b9873a

	#pragma unroll bucket_size
	for (cuda::std::int32_t i = 0; i < bucket_size; ++i) {

cuCollections/include/cuco/detail/roaring_bitmap/roaring_bitmap_impl.cuh

Lines 119 to 120 in 3b9873a

	#pragma unroll
	for (cuda::std::int32_t i = 0; i < storage_ref_.metadata().num_containers; i++) {

Describe the solution you'd like

We should leverage the new static_for in cuco to improve loops that currently rely on #pragma unroll. CCCL introduced static_for to support compile-time static loops: link. Unlike a simple pragma, it can guarantee that the loop is fully unrolled. We should apply static_for wherever the loop size is known at compile time.