* op-node: Prefer following seq drift instead of conf depth

We found a log on a testnet where a batch was rejected because it's
timestamp was greater than the origin timestamp + sequencer drift.
I looked into origin selection and determined we produce empty blocks
if the block time is ahead of the origin + seq drift; however if there
is a next origin, it is not enough to produce an empty block, we must
change the origin.

I hypothesize that this occurred because the L1 head that the node was
looking at was lagging. That would cause the node to keep using an
old origin even if the timestamp was past the max drift. I added
a unit test which shows this case & would fail without this fix
being applied.

* op-node: Don't immediately attempt to advance origin

If the current origin is past the sequencer drift, it needs to
ignore the conf depth staying, but does still need to check next
L2 block time against the next origin time.

I've also added a test for this case.