Batch remote bitfield index refreshes

[marcus-pousette-hp]

Receiving a multi-page remote bitfield currently refreshes the sparse quickbit index once per 4 KiB page. For a full 2 MiB segment, that rebuilds the same segment index 64 times.

This changes RemoteBitfield.insert() to copy all touched pages first and refresh each touched segment once. It also updates clear() to keep the sparse index correct for firstUnset() by updating the affected 128-bit quickbit index chunks.

Local microbench, 100 repeated inserts of the same remote bitfield:

bitfield size	pages	before	after
4 KiB	1	~0.04ms/insert	~0.04ms/insert
16 KiB	4	~0.27ms/insert	~0.06ms/insert
64 KiB	16	~2.06ms/insert	~0.14ms/insert
256 KiB	64	~21.0ms/insert	~0.28-0.33ms/insert

The full-segment case improves by roughly 60x because it now rebuilds the sparse quickbit index once per touched segment instead of once per 4 KiB page.

PR as co-written with AI