While checking for lost server commit holds, I noticed that the
advance_seq request path had obviously incorrect unwinding after getting
an error. Fix it up so that it always unlocks and applies its commit.
Signed-off-by: Zach Brown <zab@versity.com>