A btrfs writer fix.
Fixed: a btrfs image could hold a tree block or file data where a
superblock copy is written. A block group whose chunk covers a copy -- at
64 MiB, and at 256 GiB on a volume that large -- handed out the addresses
that map there, and the copy, written last, replaced 4 KiB of what was
allocated. Under the default dup metadata profile that damaged one copy of
one tree block in an image with more than about 27 MiB of tree blocks;
under the single profile it could replace a tree block with no other copy,
or 4 KiB of a file. A block group that covers a copy leaves the 64 KiB
stripe holding it unallocated and records it as free space, as mkfs.btrfs
does. `btrfs check --check-data-csum` reports the damage in an image 0.5.0
or 0.5.1 wrote.
The btrfs gates read every line btrfs check prints as well as its exit
status, since the checker reports a damaged copy of a mirrored block, reads
the other, and exits zero.
The public API, the command line's options, and the minimum supported Rust
version, 1.88, are unchanged.
Full notes in CHANGELOG.md.