WIP: Support for asymmetrical padding

[IanButterworth]

Based on discussion at FluxML/Flux.jl#258

[codecov-io]

Codecov Report

Merging #84 into master will increase coverage by 0.37%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master      #84      +/-   ##
==========================================
+ Coverage   70.46%   70.84%   +0.37%     
==========================================
  Files           9        9              
  Lines         579      583       +4     
==========================================
+ Hits          408      413       +5     
+ Misses        171      170       -1

Impacted Files	Coverage Δ
src/impl/conv.jl	87.58% <100%> (+0.16%)	⬆️
src/NNlib.jl	100% <0%> (+100%)	⬆️

---

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 085adb7...416cddd. Read the comment docs.

[staticfloat]

I just wanted to quickly note here that although my recently-merged im2col() rewrite has clobbered the diff here, I kept this PR in mind so adapting to the new im2col() should be very easy. Let me know if you have any questions!

[MikeInnes]

(And apologies for not getting to this earlier -- but it'd be great to get it in!)

[staticfloat]

@ianshmean Thanks for your effort here; I incorporated these changes into my recent overhaul of NNlib, and with FluxML/Flux.jl#718, we have full asymmetric padding support!

[IanButterworth]

@staticfloat Great! That makes building out YOLO a lot easier (save the loss functions, which I've not found the time to finish)