Lin bytes test fixes

[jmid]

The Lin Bytes tests are a pain point as witnessed by #544 #541 #526 #520 #518.

This PR improves on the situation by adding a Bytes.to_seq target which is a known source of domain-unsafety from the STM.domain test, yet was missing from the Lin tests. This addition means the Lin_thread test triggers much more reliably too, so it can be turned into an appropriate negative test.

Finally, shrink time of the Lin Bytes Domain test is a genuine issue #520 (it just happened again today). While trying to ramp up the targeted bindings of the corresponding STM Bytes tests today, I was playing with adding a cmd shrinker and started seeing similar longer shrink times when I added it for

 val Bytes.blit_string : string -> int -> t -> int -> int -> unit;

One problem is shrinking order: in STM, writing the shrinker by myself I could customize it to my liking, reducing the length and index integers first, and only the string secondly. Another problem is the string shrinker reducing each char repeatedly:

Shrink.string "zzzz" print_endline;;
zz
zz
zzz
nzzz
tzzz
wzzz
yzzz
znzz
ztzz
zwzz
zyzz
zznz
zztz
zzwz
zzyz
zzzn
zzzt
zzzw
zzzy

This will take quite a bit of time to reach the 'a' goal for each entry, and repeatedly start over attempting to do so.
To improve on the situation, the PR therefore adds a quick-and-dirty char-shrinker for both strings and bytes in Lin, that simply attempts to reduce a non-'a' to 'a'. As a result, shrinking for these should now work much faster out of the box.

Overall I'm hoping this fixes #544 #541 #520 #518 and thus represents a good step forward reliability-wise.
#526 is most likely due to Cygwin (signal?) issues and so remains open.

[jmid]

CI summary: all 36 workflows passed

[jmid]

CI summary for merge to main: all 37 workflows passed

[shym]

It seems that all the scheduled runs on 5.2 failed on this test this morning.
As 5.4 will enter freeze soon, maybe now is a reasonable time to retire 5.2 testing?

[jmid]

My bad, sorry! 😬 The new improved ™️ Thread mode utilizes Gc.Memprof support to trigger context switching. 5.3 restored Memprof, which means the thread test works well there and on trunk - both tested by CI on PRs.
I tested locally on 5.2 at some point and then forgot about it again. Given the above, I think the Lin Thread-mode test of Bytes should just be disabled on 5.2 and earlier.