btrfs: disable fallocate in ZONED mode

fallocate() is implemented by reserving actual extent instead of
reservations. This can result in exposing the sequential write
constraint of host-managed zoned block devices to the application, which
would break the POSIX semantic for the fallocated file.  To avoid this,
report fallocate() as not supported when in ZONED mode for now.

In the future, we may be able to implement "in-memory" fallocate() in
ZONED mode by utilizing space_info->bytes_may_use or similar, so this
returns EOPNOTSUPP.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>