[math] Tests `roottest_root_math_smatrix_testKalman` and `testOperations` fail since 1st of December 2023 on `debian10-i386`

[guitargeek]

See:
https://lcgapp-services.cern.ch/root-jenkins/view/ROOT%20Nightly/job/root-nightly-master/LABEL=ROOT-debian10-i386,SPEC=default,V=master/

The last successful build on that platform was in the morning of November 30, so any of the PRs merged on Monday seems to have broken the 32 bit build. The candidate PRs are in chronological order (newest to oldest):

• Add ROOT_CXX_STANDARD in ROOTConfig.cmake  #14148
• Allows ROOT to build with recent versions of Protobuf #14144
• [ntuple] Add more getters to RNTupleWriter #14120
• [ntuple] Throw exception when trying to load out-of-bounds entries #14129
• [ntuple] Rename R*Info classes to R*Inspector #14131
• Fix invalid memory access in TRandom3 #14135
• [ci] Remove --shm-size=1g -m 16: #14146
• [interpreter] Fix LLVM target selection #14115
• [cling] Handle variable templates in DeclUnloader #14110
• [cling] DeclUnloader: fix unloading of member functions of templated classes #13565
• Remove duplicated key from .clang-format #14137

Any PR that got backported to 6.30 is probably not the offender, because the 6.30 nightlies don't have these failures.

[guitargeek]

This problem was solved by putting the Debian 10 machine to a colder place. The timeouts where caused by CPU throttling because the machine ran to hot.

While investigating this issue, I figured out that we could also speed the tests up by using the -O3 optimization flag (as in Release mode). The RelWithDebInfo build type that is used in the nightlies implies -O2.