Optimize UFFD pager cache concurrency

[sjmiller609]

Summary

• shard the UFFD pager page cache to reduce burst contention
• add an immutable borrowed cache-hit path for UFFDIO_COPY
• expose pager timing counters for cache lookup, backing reads, page faults, and copy latency
• add a regression test for aligned page offset distribution across shards

Test Plan

• go test ./lib/uffdpager ./lib/hypervisor/firecracker
• remote browser burst benchmarks on deft-kernel-dev with UFFD enabled

[sjmiller609]

Closing in favor of #262, which folds this UFFD pager cache concurrency work into the new UFFD pager PR. Keeping the old branch around for comparison.