[WIP] Sparse semiring follow-on & optimizations

[cjnolet]

Closes #3371

A couple changes requested in #3157 that should be applied to this PR:

• Explicit gtests for coo_spmv and csr_spmv
• remove device allocator, stream, and cusparse handle from distances_config_t and just use the raft handle
• enable smem to store vectors from A in sparse form in coo_spmv (this is being investigated in [WIP] Sparse distances hash table #3479)
• try different optimizations outlined in [FEA] Optimizations to sparse semiring distances #3371

[cjnolet]

Current plan is to optimize where at all possible:

1. when n_cols < smem available: use coo_spmv w/ smem to store A in dense form
2. when n_cols > smem available but max_degree(A) < smem available: use coo_spmv w/ hash table in smem to store A in sparse form (and revert to binary search in smem when the hash table becomes full enough).
3. when n_cols > smem available and max_degree(A) > smem available but only a few entries have that max degree and the rest are closer to 0, use chunks (multiple blocks) of coo_spmv w/ chunks over only rows w/ large degrees to store A in sparse form in smem.

Either 1 is able to be performed, or 2 & 3 can be performed independently. While different amounts of smem could be required (and less smem used will allow more blocks to be scheduled), it might be better to launch different kernels for each of the strategies, based on how much smem would be used. Perhaps this could be done by diving the rows of A into buckets and launching separate kernels (one for each bucket).

[ajschmidt8]

@cjnolet, can you try updating your branch? The PR below just merged which should get branch-0.19 up-to-date and remove some of these extraneous commits from your PR.

#3444

[github-actions]

This PR has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this PR if it is no longer required. Otherwise, please respond with a comment indicating any updates. This PR will be labeled inactive-90d if there is no activity in the next 60 days.