STM Weak test re-revision

[jmid]

I recently had a fresh look at the weak array and weak hash set STM tests, and discovered a couple of things.
This PR is the outcome.

The weak array and weak hash set test originated in #214.
They've since been tweaked, e.g., in #365 which among other things switched to int64 data for the weak hash sets test,
arguing that this was faster to shrink. As we no longer shrink the data this argument no longer holds.
Furthermore, the tests triggered crashes in the past with string data in #283, which was caused by a combination of race conditions and variable length memory records. This speaks for reverting and using strings for both tests. I've confirmed that both parallel tests can trigger segfaults on an older version (OCaml 5.0.0~alpha1) with these changes.

The PR also

• revises the post-conditions slightly,
• adjusts the frequencies slightly to skew them towards making additions, and
• cleans up the protect calls with thunks, rather than the current missing-last-argument-closure

[jmid]

CI summary for 7ce003b: all 13 trunk workflows (except trunk debug) failed due to #559

• 32bit trunk
• Bytecode trunk
• Cygwin trunk
• FP trunk
• Linux trunk
• Linux-ARM64 trunk
• macOS-ARM64 trunk
• macOS-intel trunk
• MinGW bytecode trunk
• MinGW trunk
• MSVC bytecode trunk
• MSVC trunk
• musl trunk

Out of 50 workflows, 13 failed with a genuine issue