[SwayamInSync]

Plan item T6 (stability CI), for exit criterion 3 (steady verdicts): nothing measured whether the Verify suite gives the same results run after run, and flaky tests showed up only as red marks on unrelated pull requests (confirmation_invariants.cpp on #105 and #110).

With this PR

cppverify-stability.yml runs every night (04:17 UTC) and on demand:

• Linux (ubuntu-24.04, the release configuration with cvc5 1.4.2, ccache from CI's cache) and macOS (macos-14, setup.sh, as CI's macOS job);
• the Verify suite five times on each, at the parallelism CI uses, each run's results kept as an artifact (lit -o);
• cppverify-stability.py compares the runs and fails the job when a test is not PASS, UNSUPPORTED, or XFAIL in some run, naming it with its result in each run, as flaky (results differ) or failing (the same failure every time):

2 tests, 3 runs: 1 flaky, 0 failing in every run

Flaky:
  Clang :: Verify/dump_ir.cpp: PASS PASS FAIL

(the script on two real runs of two tests and a third with one result changed by hand).

Seeds are not part of it: cpp-verify does not expose the solver's random seed, and a test that checks a counterexample's values would change with it although its verdict does not.

Not tested here

A workflow can be started by hand only once it is on the default branch, so its first run will be after the merge; it reuses the build steps of cppverify-differential.yml (Linux) and CI's macOS job unchanged.