Remove NaN and Inf checks if __FINITE_MATH_ONLY__ is defined

[garrison]

This is motivated by qiskit-community/fulqrum#3. If finite-only math is used, then these checks are unnecessary, and we can remove them to spare the user some warnings.

However, I am hesitant to go out of my way much to support compilers that have been instructed to break IEEE 754 semantics, so I am still considering whether it makes sense to merge this.

• Should we add tests in CI with fast-math enabled? This would imply we support this explicitly, which may be a bad idea.
• Should we instead consider hard-failing if __FAST_MATH__ is defined?

[coveralls]

Pull Request Test Coverage Report for Build 23369575181

Details

• 0 of 0 changed or added relevant lines in 0 files are covered.
• 19 unchanged lines in 5 files lost coverage.
• Overall coverage increased (+6.4%) to 57.387%

---

Files with Coverage Reduction	New Missed Lines	%
include/qiskit/addon/sqd/internal/sample-without-replacement.hpp	1	67.8%
include/qiskit/addon/sqd/configuration_recovery.hpp	2	80.0%
include/qiskit/addon/sqd/subsampling.hpp	4	64.0%
test/test_configuration_recovery.cpp	5	50.78%
include/qiskit/addon/sqd/fermion.hpp	7	53.57%

Totals
Change from base Build 20271294895:	6.4%
Covered Lines:	377
Relevant Lines:	408

---

💛 - Coveralls

[garrison]

After further thought, I think we should disable these checks if __FINITE_MATH_ONLY__ is defined, as this PR does. We should also display a warning if a user compiles with __FAST_MATH__ defined, and we should document that we do not support the use of fast math.