tests/intel/gem_exec_capture: changing the test size breaks the run

After fixing most of the tests/intel/gem_exec_capture: many-* subtests with this patch, I noticed a few oddities.

Sub-issue 1: tests fail if called in a sequence:

1. The tests will run fine when running one at a time with --run-subtest, but only the first one will succeed when they are run in a sequence such as when calling ./builddir/tests/gem_exec_capture without arguments.
2. When running multiple tests in a sequence, changing the size argument of the function many() is what seems to cause the failure.
3. If a test fails, calling it a second time solves the issue. Calling the same test multiple times in a sequence will also work.
4. Closing and reopening the file descriptor fd between tests solves the issue

Sub-issue 2: Capturing 0 blobs:

1. On success, the test will print the number of blobs captured. For example, for the many-2M-incremental, the output may be Captured 0 512-blobs out of a total of 256.
2. Running the same test again without closing and reopening the fd will change the number of captured blobs to max, or to 256 in our example: Captured 256 512-blobs out of a total of 256.

My first idea was that there was some memory content leftover from the previous test that was causing these issues(assuming they are related), but I found no evidence of that.

Do you have ideas I can explore to try to find out what is going on here?