Include discussion of performance vs. accuracy tradeoffs and comparison to fastmath in docs

[danielwe]

This draft PR is based on a recent slack thread about the arithmetic transformations and function substitutions performed by @turbo and how they affect both accuracy and performance, as well as how they compare to the transformations performed by @fastmath. Thanks to @chriselrod and @oscardssmith for the incredibly helpful discussion. It would be great to preserve these insights and make them easy to find for future users. For now, I've copied the discussion verbatim over to a separate page in the docs. Aiming to come back and edit this into prose soon, but maintainers are also welcome to edit freely.

[codecov]

Codecov Report

Merging #389 (72edb2f) into master (23f5d56) will increase coverage by 2.36%.
The diff coverage is n/a.

❗ Current head 72edb2f differs from pull request most recent head 2a37fd7. Consider uploading reports for the commit 2a37fd7 to get more accurate results

@@            Coverage Diff             @@
##           master     #389      +/-   ##
==========================================
+ Coverage   84.76%   87.12%   +2.36%     
==========================================
  Files          38       38              
  Lines        9417     9437      +20     
==========================================
+ Hits         7982     8222     +240     
+ Misses       1435     1215     -220     

Impacted Files	Coverage Δ
src/codegen/lower_threads.jl	49.04% <0.00%> (-9.99%)	⬇️
src/simdfunctionals/mapreduce.jl	91.95% <0.00%> (-0.52%)	⬇️
src/simdfunctionals/filter.jl	89.65% <0.00%> (-0.35%)	⬇️
src/precompile.jl	100.00% <0.00%> (ø)
src/predicates.jl	100.00% <0.00%> (ø)
src/constructors.jl	98.23% <0.00%> (ø)
src/getconstindexes.jl	100.00% <0.00%> (ø)
src/parse/add_loads.jl	98.57% <0.00%> (ø)
src/LoopVectorization.jl	75.00% <0.00%> (ø)
src/codegen/split_loops.jl	100.00% <0.00%> (ø)
... and 28 more

---

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 5d6f138...2a37fd7. Read the comment docs.

[chriselrod]

Are you planning on getting back to this?
If you don't want to edit the content, you could just make a comment that this was a slack discussion, and that anyone reading it is welcome to make a PR to turn the conversation into a more concise explanation.

[danielwe]

Yeah, realistically it would probably be a while before I'd get around to editing this. How does it look now?

[chriselrod]

Yeah, I think this is fine, thanks!