Reduce STM.cmds_gen list size

[jmid]

STM_sequential uses Gen.sized to generate cmd lists, which means the output size depends on Gen.nat, which may output lists of up size 10.000 with a 5% chance.

Here's are some quick stats to illustrate, here on the Domain.DLS STM test:

multicoretests-latest$ dune exec src/domain/stm_tests_dls.exe -- -v
random seed: 245423950
generated error fail pass / total     time test name
[✓] 1000    0    0 1000 / 1000     0.5s STM Domain.DLS test sequential

+++ Stats for STM Domain.DLS test sequential ++++++++++++++++++++++++++++++++++++++++++++++++++++++++

stats cmd length:
  num: 1000, avg: 364.61, stddev: 1267.28, median 9, min 0, max 9737
     0.. 486: #######################################################         856
   487.. 973: ######                                                           94
   974..1460:                                                                   4
  1461..1947:                                                                   2
  1948..2434:                                                                   5
  2435..2921:                                                                   4
  2922..3408:                                                                   2
  3409..3895:                                                                   2
  3896..4382:                                                                   2
  4383..4869:                                                                   4
  4870..5356:                                                                   2
  5357..5843:                                                                   6
  5844..6330:                                                                   0
  6331..6817:                                                                   2
  6818..7304:                                                                   3
  7305..7791:                                                                   2
  7792..8278:                                                                   0
  8279..8765:                                                                   1
  8766..9252:                                                                   4
  9253..9739:                                                                   5
================================================================================
success (ran 1 tests)

A cmd list of length 9737 is excessive - and now hurts client users of the library, such as Ortac's QCheck-STM plugin!

This PR therefore proposes to replace the distribution with an exponential distribution instead.

For a start I've gone with a mean of 10, and added a bit of skew to avoid generating too many empty cmd lists, which should be less interesting in a state-machine setup. The resulting distribution looks as follows (now with count raised to 10000, and a 230/10000 ~ 2.3% chance of generating empty cmd lists which seems reasonable):

multicoretests-latest$ dune exec src/domain/stm_tests_dls.exe -- -v
random seed: 7625658
generated error  fail  pass / total     time test name
[✓] 10000     0     0 10000 / 10000    15.3s STM Domain.DLS test sequential

+++ Collect ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Collect results for test STM Domain.DLS test sequential:

not mt: 9770 cases
empty : 230 cases

+++ Stats for STM Domain.DLS test sequential ++++++++++++++++++++++++++++++++++++++++++++++++++++++++

stats cmd length:
  num: 10000, avg: 10.27, stddev: 9.95, median 7, min 0, max 79
   0.. 3: #######################################################        2749
   4.. 7: ################################################               2417
   8..11: ###############################                                1568
  12..15: #####################                                          1095
  16..19: ##############                                                  706
  20..23: #########                                                       453
  24..27: ######                                                          314
  28..31: ####                                                            248
  32..35: ##                                                              147
  36..39: ##                                                              116
  40..43: #                                                                62
  44..47:                                                                  39
  48..51:                                                                  24
  52..55:                                                                  28
  56..59:                                                                  14
  60..63:                                                                   9
  64..67:                                                                   4
  68..71:                                                                   3
  72..75:                                                                   2
  76..79:                                                                   2
================================================================================
success (ran 1 tests)

I'm curious to see how this fares on the CI.

Shoutout to @nikolaushuber for reporting this.

Note to self: might warrant a changelog entry.

[jmid]

CI summary for d950ed0: all 45 workflows completed succesfully!

[jmid]

CI summary for 65edac1: all 45 workflows completed succesfully!

[jmid]

CI summary for a661f27: all 45 workflows completed succesfully!

Merging...

[jmid]

CI summary for merge to main:

• linux-s390x-5.2 failed with a timeout s390x timeouts on s390x-worker-01 #421

Out of 46 workflows 1 failed with a CI issue