Fix generic matmult in LinearAlgebra.jl

[odow]

Closes #175

[codecov]

Codecov Report

Base: 86.14% // Head: 86.17% // Increases project coverage by +0.02% 🎉

Coverage data is based on head (9b7ee8d) compared to base (352aa8f).
Patch coverage: 100.00% of modified lines in pull request are covered.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #176      +/-   ##
==========================================
+ Coverage   86.14%   86.17%   +0.02%     
==========================================
  Files          20       20              
  Lines        1877     1881       +4     
==========================================
+ Hits         1617     1621       +4     
  Misses        260      260              

Impacted Files	Coverage Δ
src/implementations/LinearAlgebra.jl	97.31% <100.00%> (+0.07%)	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[odow]

This whole system of LinearAlgebra with abstract types is a mess. Base fails, MA fails, and there are changes between Julia versions :(

[blegat]

Yes, I think this fix is the right thing to do. We don't want to look at the first elements and broaden the array type if needed etc... If the eltypes are not concrete it's ok to use a less performant approach.
This could cause issues with users suddenly getting extremely bad performance but then I think we could add an additional preprocessing here that tries to see if all elements are actually of the same type even if the eltype is not concrete.