don't specialize any functions

[simeonschaub]

This takes the time to run the ChainRules test suite from 52 minutes
down to 40 in my tests. I think there are still more gains to be had
here by playing around a bit with Experimental.@compiler_options, but
initially that caused some of the inference tests to fail.

[codecov-commenter]

Codecov Report

Merging #205 (b16b0e7) into master (011b4e2) will increase coverage by 0.06%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #205      +/-   ##
==========================================
+ Coverage   90.94%   91.00%   +0.06%     
==========================================
  Files          11       11              
  Lines         287      289       +2     
==========================================
+ Hits          261      263       +2     
  Misses         26       26              

Impacted Files	Coverage Δ
src/ChainRulesTestUtils.jl	100.00% <ø> (ø)
src/iterator.jl	96.29% <0.00%> (+0.29%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 011b4e2...b16b0e7. Read the comment docs.

[oxinabox]

This makes sense.
So many of the things we do, we do only once.
Since we are testing a particular type combo.

What happenes if we turn it back on for the rand_tangent file?
Those methods we do call a lot with the same types

[simeonschaub]

Good point! Trying that right now.

[simeonschaub]

What happenes if we turn it back on for the rand_tangent file?
Those methods we do call a lot with the same types

Looks like that helps in some tests, but was overall slower. We probably don't spend enough time in rand_tangent that this specialization is worth it.