[test] make fixarr3 and fixarr2 non-concurrent

[ferdymercury]

Fixes #19076

This way they are executed one after the other, as they were in the original Makefile.

One hypothesis is that, even in "READ" mode, same ROOT file accessed concurrently might interfere. see eg #18954
due to TFile::SysClose or so.

Not sure if an alternative approach would be to read without global registration or so?

[linev]

It would be really strange if concurrent reading makes problems

[ferdymercury]

You mentioned that if failed 10% of the time on your Mac. Does the percentage change after adding this fix ?
Right now mac13, mac15 and mac15 went well.

It would be really strange if concurrent reading makes problems

Yeah, weird. Maybe the problem is the "delete" part after GetObject?

[linev]

Last days I see no any failures for these tests.

And any GetObject problems cannot affect other process which only reads same file.

[ferdymercury]

Last days I see no any failures for these tests.

even after reenabling fixarr2 ?

I am exploring that in the CI, see #19762

[github-actions]

Test Results

    18 files      18 suites   2d 20h 38m 19s ⏱️
 3 635 tests  3 460 ✅  0 💤 175 ❌
64 024 runs  63 825 ✅ 23 💤 176 ❌

For more details on these failures, see this check.

Results for commit f2c571f.