Run MMQA tests as part of CI

[rptb1]

Attempting to resolve #106

[rptb1]

This is currently failing to pass test function/136.c https://github.com/Ravenbrook/mps/actions/runs/3939501021/jobs/6739430364#step:6:865 . Running this locally, I find an assertion failure here

mps/test/function/136.c

Lines 81 to 86 in ed73ba8

	/* Allocate one small object in pool2 so that its CBS gets some
	initial memory. */
	res = mps_alloc(&p, pool2, 8);
	asserts(res == MPS_RES_OK,
	"Couldn't allocate first object of size %lu in second pool",
	(unsigned long)8);

The test reduces the arena commit limit, but seems to assume that it will be able to allocate an 8-byte object in pool2. This call fails with result code 7, which means it's hitting the commit limit. The test appears to expect to fail to allocate, but only for the "large objects" that it tries next. This seems wrong to me, especially if the commit limit is reached exactly by allocating the descriptors for the pool creations.

[rptb1]

Possibly related to #88 but trying function/136.c from that branch does not seem to fix the problem here.

EDIT: Disregard this. I needed to clean.

[rptb1]

This is currently failing to pass test function/136.c

There seems to be a sensitive boundary condition. Running

rb@plover:~/git/mps/test$ perl test/qa run function/136.c
...
RESULTS
  assert             =  true
  assert_or_abort    =  true
  asserttext         =
    Couldn't allocate first object of size 8 in second pool
  limit_grains       =  39
  seconds            =  0
CONCLUSION
  PASS

but running

rb@plover:~/git/mps/test$ perl test/qa runset testsets/passing
...
  PASS function/135.c (writeable)
X FAIL function/136.c (writeable)
  PASS function/137.c (writeable)
...

with this in log/full.log:

RESULTS
  assert             =  true
  assert_or_abort    =  true
  asserttext         =
    Couldn't allocate first object of size 8 in second pool
  limit_grains       =  40
  seconds            =  0
CONCLUSION
  FAIL                  failed on limit_grains: wanted < 40, was 40

So it seems that test/qa run and test/qa runset are not equivalent.