Specialize foldl for cartesian style arrays

[tkf]

Commit Message

Specialize foldl for cartesian style arrays (#407)

This patch implements a specialization of foldl on arrays with
IndexCartesian index style. This gives us more than 2x speedup (see
sum_transpose benchmark).

The implementation mostly just redirects the call to foldl of
CartesianIndices. Most of the code is for compatibility with Julia
< 1.5. This PR also fixes a bug in foldl for multi-dimensional
Broadcasted (probably introduced by #403).

[codecov]

Codecov Report

Merging #407 into master will increase coverage by 0.06%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #407      +/-   ##
==========================================
+ Coverage   90.82%   90.89%   +0.06%     
==========================================
  Files          25       25              
  Lines        1581     1592      +11     
==========================================
+ Hits         1436     1447      +11     
  Misses        145      145              

Impacted Files	Coverage Δ
src/basics.jl	92.10% <100.00%> (+1.48%)	⬆️
src/processes.jl	93.03% <100.00%> (+0.17%)	⬆️

---

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 b8efeed...202df0a. Read the comment docs.

[tkf]

https://github.com/JuliaFolds/Transducers-data/blob/benchmark-results/2020/08/09/100019/result.md

ID	time ratio	memory ratio
["sum_transpose", "30", "noinit", "xf"]	0.43 (5%) ✅	Inf (1%) ❌
["sum_transpose", "30", "withinit", "xf"]	0.41 (5%) ✅	Inf (1%) ❌