f24e4df3

By: ValentinHirschi <valentin.hirschi@gmail.com>

Share the checkpoint manifest byte cap with the shard supervisor

Review findings #2/#13: the walker reads `latest.json` under a 16 MiB cap
while the shard supervisor kept the old 1 MiB bound, so a valid paused
child with a larger manifest would have been classified as failed and
halted the campaign. The store's `MAX_MANIFEST_BYTES` is exported as
`OWNER_DOMAIN_WALK_CHECKPOINT_MANIFEST_MAX_BYTES` and the supervisor uses
it; the fake child in the supervisor tests now writes a manifest above
1 MiB so the pause path exercises the shared bound. Segment-list
compaction (the manifest grows about three entries per generation) is
recorded as a wave-2 follow-up in the CP5 note.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>