The special handling of PF_MEMALLOC callers in ext4_write_inode()
shouldn't be necessary as there shouldn't be any. Warn about it. Also
update comment before the function as it seems somewhat outdated.

(Changes modeled on an ext3 patch posted by Jan Kara to the linux-ext4
mailing list on Februaryt 28, 2014, which apparently never went into
the ext3 tree.)
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: Jan Kara <jack@suse.cz>