btrfs: dev-replace: count write errors to the target again
btrfs_dev_replace::num_write_errors is reported by the dev-replace status
ioctl, and scrub_enumerate_chunks() stops a replace with -EIO once it is
non-zero, so that a target that fails writes does not silently end up as
the new device. Since the dev-replace writeback moved to the
scrub_stripe infrastructure nothing increments it: status always shows 0
write errors, and the abort never happens.
After the replace writes of a stripe have completed, add the sectors
whose write failed. The stripe's write_error_bitmap also collects failed
repair writes to the source device, so clear it before the replace
writes are submitted.