Test `SELECT FOR UPDATE` for CRDB overlap
When we initially implemented overlap keys, SELECT FOR UPDATE didn't suffice to make writes conflict. There was some language in more recent CRDB releases that suggest it may work now. It's worth updating the e2e test suite with new CRDB versions and testing SELECT FOR UPDATE instead.
@ecordell can you please clarify why SELECT FOR UPDATE didn't suffice to cause a conflict?
There's some other discussion here: https://github.com/authzed/spicedb/issues/2241#issuecomment-2652388873
Basically the test suite proved that select for update could still get reversed timestamps. My full notes from the experiment are here: https://hackmd.io/88_ybweDS0m7lKLj_0uqHw?view