Make Buffer track per-block (not per-byte) ownership

[mkeeter]

This reduces storage (and operations) in the struct Buffer by the block size, so a factor of 512× or 4096×.

The downside is that we need to know the block size at construction time, so there's a bunch of mechanical changes in this PR.

I see significant speedups for all sizes of random reads except 4K, which become slightly slower (50.6 MiB/s → 46.9 MiB/s, not sure why).

[mkeeter]

Re-running on Crucible main (099c201), I see 44.5 MiB/s, so I suspect the 50.6 MiB/s reading was a fluke. This means that there's no performance regression, so we can move forward with the PR.

[mkeeter]

(let's put this on hold until #1094, which touches a bunch of similar code)

[mkeeter]

Closing in favor of #1105, since this has diverged from main and I want to split into two bite-sized PRs.