Reenable Bigarray STM tests

[jmid]

This PR reenables the Bigarray STM tests, the idea being that a model-based STM test is stricter than a Lin one.

A bit of background:

• Lin tests of Bigarray were added in #121
• STM tests of Bigarray were added in #125
• The STM Bigarray tests were disabled in #250 (by mistake AFAICS)
• This PR leaves stress tests of Bigarray using the Lin spec #443

[jmid]

The two 32-bit runs are failing since size_in_bytes return 4b * 16 = 64b and breaks with the 8b word assumption hardcoded in the mode. I guess #250 predates us testing 32-bit... :shrug:

On macOS 5.2 and MSVC bytecode the parallel STM Bigarray tests failed to trigger a counterexample within 1000 iterations. The old Lin test had a count of 5000 for this reason.

CI summary for 4713ced:

• 32bit 5.2 fails due to the above size_in_bytes model mismatch
• 32bit trunk fails due to the above size_in_bytes model mismatch
• MSVC bytecode trunk failed to find a parallel STM Bigarray counterexample within 1000 iterations.
• MSVC trunk crashed in Lin Dynlink stress test with Domain #307
• MinGW 5.2 timed out in Lin Dynlink stress test with Domain #307
• MinGW trunk timed out in Lin Dynlink stress test with Domain #307
• macOS 5.2 failed to find a parallel STM Bigarray counterexample within 1000 iterations.

[jmid]

Rebased on main after the merge of #453 to remove noise.

• a3f7202 should take care of the 32 failures
• e9421bb should take care of the failures to trigger a counterexample within 1000 iterations

[jmid]

CI summary:

• Cygwin 5.2 timed out in negative Lin Dynlink test with Domain #307

This one was missed in #453 - #454 addresses the omission

Out of 37 workflows 1 failed with a genuine issue

[jmid]

CI summary for merge to main:

• Cygwin 5.2 timed out in negative Lin Dynlink test with Domain #307

Out of 38 workflows 1 failed with a genuine issue