FCBIS mixed-version sync on macOS crashed in recovery with EINVAL after salvaging leftover bytes in a preallocated journal tail. The leftover-tail NOTICE is longstanding on macOS; this ticket is the subsequent recovery failure after salvage.

Seen in BF-45282: fcbis_mixed_version_sync.js on enterprise-macos. An FCBIS node on latest FCV copied files from a last-LTS replica set. The FCV downgrade close/reopen then failed:

journal/WiredTigerLog.0000000002: log record at position 33037184 has record len corruption 0x0
log file journal/WiredTigerLog.0000000002 corrupted at position 33037184, truncated.
log file journal/WiredTigerLog.0000000003 corrupted at position 69933952, truncated.
int __wt_txn_recover(...):1462:Recovery failed  error_str: Invalid argument  error_code:22

MongoDB leaves log.zero_fill off, so journal preallocation uses __wt_fextend. Linux fallocate zeros that tail. macOS has no fallocate and falls back to ftruncate, which is sparse and can leave leftover non-zero bytes in the unused 100MB tail. Recovery reports those as record len corruption 0x0.

The first reopen after FCBIS is a backup restore (WT_CONN_WAS_BACKUP) and _log_has_hole skips record verification. The extra close/reopen from the FCV downgrade does verify, salvages by truncating at the last good record (log_truncate at rd_lsn), then recovery still fails with EINVAL. _log_salvage_message reports bad_offset (leftover in the preallocation); the truncate point is rd_lsn.

Same NOTICE: WT-4233, WT-14850, WT-17183. Those salvage and continue. This one fails because the extra reopen after backup restore hits EINVAL.

Need to truncate so a reopen after salvage does not re-read the leftover tail and fail.