Add support for ebpf_hash_table_t with preallocated buckets

Up to 50% of map update is due to pool allocations.

It might be possible to reduce this if we preallocate the buckets.

Algorithm:
1) Allocate an array of 1 entry buckets.
2) Each entry in the array has a released epoch.
3) If released epoch == UINT64_MAX, then the bucket is busy.
4) If (bucket's release epoch <= global release epoch), then it can be reused, mark release epoch == UINT64_MAX.
5) When freeing a bucket, if it's the preallocated bucket, set release epoch == global current epoch.

Buckets can then be in 3 states:
1) In use - release epoch == UINT64_MAX.
2) Waiting to be reused - release epoch > global release epoch.
3) Ready to be reused - release epoch <= global release epoch.

Start by tracing the map update path and the pool allocation work described in the issue. Assess how preallocated one-entry buckets would interact with release epochs, including the in-use, waiting, and reusable states. Done means map updates avoid eligible pool allocations while preserving safe bucket reuse.

Written by the indexing model from the issue text.