This is a bit awkward: we're passing around a btree_trans, but we're not
in a context where transaction restarts are handled - we should try to
come up with a better way to denote situations like this.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>