Fix rep_count in STM_thread

[jmid]

I accidentally discovered that the two rep_counts in STM_thread were different:

• agree_test_conc would perform 100 repetitions whereas
• neg_agree_test_conc would perform only 25 repetitions

I think this is simply a copy-paste error.

I wonder whether this would allow us to trigger STM_thread test errors more consistently... 🤔

[jmid]

Rebased on main and added a CHANGES entry

[jmid]

CI summary:

• The Out_channel test failed to trigger on Windows bytecode and macOS trunk
  • it also took 9440.3s on macOS causing a time out and appears to cause a timeout on macOS trunk
• Cygwin trunk part 1 timed out with a number of different tests taking a bit long
• Lin DSL Weak HashSet test took an extraordinary 14177.8s to shrink on Windows bytecode 5.0.0 and 9009.0s on Cygwin trunk part 2 both causing a timeouts
• threadomain dead locked on Windows trunk

None of these are related to this PR so I'll merge.