Ephemeron test revision

[jmid]

With #365 merged this PR offers a similar revision of the Ephemeron tests, where we have observed excessive shrinking runs, primarily under Cygwin where they have contributed to workflow timeouts.

Similarly to the Weak module, Ephemerons depend on the GC heap state - and so we avoid Lin tests that compare the state of potentially widely different heap states in favor of STM tests, taking extra care

• to Gc.full_major before starting the parallel test,
• to run a Gc.minor in between, and
• to actually use boxed types

With luck, we should hopefully see fewer Cygwin timeouts due to excessive Weak and Ephemeron shrinking runs

[jmid]

CI summary for 67465a2:

• Cygwin part 1, 5.1~alpha2 and trunk 5.2 timeouts with +1000sec runs of Domain.spawn/join - atomic, domain_spawntree, Lin Atomic, Bigarray, and Bytes
• Cygwin part 1, trunk 5.2 timeout with the above + a 2795.8s Lin Out_channel shrinking run
• Windows bytecode trunk 5.2 and Windows 5.1~alpha2 timeout on threadomain [ocaml5-issue] Windows failures on threadomain #203

[jmid]

CI summary for 4907ffa:

• Windows 5.1~alpha2 failure to trigger STM int ref test parallel asymmetric 'STM _ ref test parallel asymmetric' failure to trigger #364
• Cygwin part 1 trunk 5.2 timeout due to +1000s Lin tests like above + 4453.6s domain_spawntree - with Atomic + 3836.7s Lin Out_channel shrink run
• Cygwin part 2 trunk 5.2 -timeout with a 7763.8s Lin ref int64 test with Thread shrinker run
• Linux arm64 5.1 and 5.2 failed to trigger Lin Out_channel failure
• Linux s390x 5.1~alpha2 unexpectedly found a Lin CList int test with Thread counterexample Lin.Internal Thread int CList test finds unexpected counterexample under bytecode #358

[jmid]

I'll merge as none of the above are related to the Ephemeron test changes