Chore: Collect magic constants

[jmid]

Hacking on #540 I realized that the {Lin,STM}_{domain,thread,...} modes contain a number of hard-earned magic constants making these modes work reasonably well.

The constants are however

• occurring directly in the code, rather than up front and
• duplicated

The PR therefore pulls them out as documented top-level definitions.
In Lin, it is a bit more messy than in STM, as the constants are spread across both lin.ml and lin_domain.ml, ...
In lin_effect.ml I also decided to remove the useless Util.repeat 1.
None of these changes should be user-visible however.

[jmid]

CI summary: all 36 workflows passed. Merging.

[jmid]

CI summary for merge to main

• MSVC bytecode trunk failed to find a Lin Bytes test with Domain counterexample Test "Lin Bytes test with Domain" is flaky #541
• Cygwin 5.3 timed out after 8667.0s shrinking in Lin Bytes test with Domain Timeout due to excessive shrinking in Lin Bytes test with Domain on Cygwin #520

Out of 37 workflows 2 failed due to false alarms (from the same test even!)