Teach julia that `NaNMath` functions don't have effects

[oscardssmith]

This allows constant folding and dead code elimination on Julia 1.8 and up.

[codecov[bot]]

Codecov Report

Patch coverage: 80.00% and project coverage change: +0.02% :tada:

Comparison is base (d5ea765) 97.05% compared to head (d51c509) 97.08%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #67      +/-   ##
==========================================
+ Coverage   97.05%   97.08%   +0.02%     
==========================================
  Files           1        1              
  Lines         102      103       +1     
==========================================
+ Hits           99      100       +1     
  Misses          3        3              

Files Changed	Coverage Δ
src/NaNMath.jl	97.08% <80.00%> (+0.02%)	:arrow_up:

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[ViralBShah]

I'm not the right person to review this.

[mlubin]

This needs a review from someone who understands @assume_effects

[oscardssmith]

I've talked to @vtjnash about this a bit and he thinks that libm might be leaking effectful information (rounding modes and signals) but I haven't been able to find observe this.

[vtjnash]

In particular, note that C documentation for these explicitly states these are not pure due to side effects.

[oscardssmith]

however the two areas of impurity (rounding modes and flags) appear not to be accessible from julia (or at least I haven't been able to observe the impurity)