opt: reduce cost of locality optimized anti join

[rytaft]

This commit reduces the cost of locality optimized anti join to
to ensure that we always choose it over a non-locality-optimized anti
join when possible.

Informs #63735

Release note (performance improvement): Adjusted the estimated cost
of locality optimized anti joins in the optimizer so that they are
always chosen over non-locality-optimized anti joins when possible.
This makes it more likely that queries involving anti joins (such as
inserts with foreign key checks) can avoid visting remote regions.
This results in lower latency.

[cockroach-teamcity]

This change is 

[RaduBerinde]

Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @rytaft)

---

pkg/sql/opt/xform/coster.go, line 881 at r1 (raw file):

	// based on the latency between regions.
	if localityOptimized {
		cost /= 2.5

Any problem with using a bigger constant (given that 2 was too conservative)?

[rytaft]

TFTR!

Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @RaduBerinde)

---

pkg/sql/opt/xform/coster.go, line 881 at r1 (raw file):

Previously, RaduBerinde wrote…

Any problem with using a bigger constant (given that 2 was too conservative)?

I don't think there's a big problem, but I also don't want the cost to be way smaller than it would be otherwise. I'm pretty confident that 2.5 will do the trick since the cost of all the lookup joins should be basically identical. I think the problem with dividing by 2 was that depending on the numerical rounding, either the locality optimized anti joins or the single anti join could be slightly cheaper.

[rytaft]

bors r+

[craig]

Build succeeded:

• GitHub CI (Cockroach)