fix IEEE compliancy in floating point min/max reductions on LLVM verson that supports it

[KristofferC]

Fixes #129

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 88.86%. Comparing base (48a8a4e) to head (41139b4).
Report is 1 commits behind head on master.

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #130      +/-   ##
==========================================
+ Coverage   88.49%   88.86%   +0.37%     
==========================================
  Files           5        5              
  Lines         565      566       +1     
==========================================
+ Hits          500      503       +3     
+ Misses         65       63       -2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[eschnett]

IEEE actually defines a "minimum" reduction that ignores nans. (I know, I know...) I would thus argue with "make min behave the same as everywhere else in Julia", not with IEEE compliance. (I agree with the change in this PR.)

[KristofferC]

I would thus argue with "make min behave the same as everywhere else in Julia", not with IEEE compliance.

Yes, I agree (but that is also what this PR should do).