opt: M1 Macs produce different result for normalization rule cycle tests

[mgartner]

On M1 Macs, this test has a different output:

https://github.com/cockroachdb/cockroach/blob/5ffe3c531385bbf52e5a414addbc489279f3d535/pkg/sql/opt/norm/testdata/rules/cycle

  # This test ensures that a rule cycle does not cause a stack overflow. In test
  # builds, a deep constructor call stack is detected and the Factory panics.
  exprnorm
  (NormCycleTestRel (True))
  ----
- error: optimizer factory constructor call stack exceeded max depth of 10000
+ values
+  ├── cardinality: [0 - 0]
+  └── key: ()

I'm not exactly sure how this test is executed differently on M1 Macs. Is some rule firing before NormCycleTestRel that prevents the infinite recursion from starting? Is the M1 somehow recognizing the infinite recursion and arbitrarily halting it?

See: https://reviewable.io/reviews/cockroachdb/cockroach/82877#-N5av5HC3Gp5QuRAfQAL

Jira issue: CRDB-17115

[mgartner]

@healthy-pod confirmed that this is no longer an issue. Closing.