Nospecialize on dynamic pairs

[Tokazama]

There shouldn't be any changes to code inference. This does more of what #13 does but focuses on methods where one argument is static and one is dynamic, making it unnecessary to specialize on the static type.

@chriselrod, if this doesn't affect performance or constant propagation for your code libraries then it should be good to go.

[codecov]

Codecov Report

Merging #19 (471e6bb) into master (baa9ee4) will increase coverage by 0.03%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master      #19      +/-   ##
==========================================
+ Coverage   98.33%   98.36%   +0.03%     
==========================================
  Files           7        7              
  Lines         421      429       +8     
==========================================
+ Hits          414      422       +8     
  Misses          7        7              

Impacted Files	Coverage Δ
src/Static.jl	97.77% <100.00%> (ø)
src/bool.jl	100.00% <100.00%> (ø)
src/float.jl	98.64% <100.00%> (ø)
src/int.jl	100.00% <100.00%> (ø)
src/symbol.jl	100.00% <100.00%> (ø)
src/tuples.jl	100.00% <0.00%> (ø)
src/ndindex.jl	93.50% <0.00%> (+0.45%)	⬆️

---

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 baa9ee4...471e6bb. Read the comment docs.

[chriselrod]

I'd have to run tests and benchmarks, but it looks like these are covered by @inferred tests here.

[Tokazama]

I'd have to run tests and benchmarks, but it looks like these are covered by @inferred tests here.

I'm fine with whatever you're comfortable with. I don't want to give you unnecessary work but I also don't want to mess up any of your code.

I think the most likely problem would be slow downs due to using @inbounds(getfield(x, :parameters)[1]) to avoid specializing on dynamic results. I don't think any static computations should be affected.

[chriselrod]

I don't really know anything about what @nospecialize does, how it is implemented, or what guarantees it'll have in the future, so I'm not really qualified to review.

[Tokazama]

All of this was implement in #55