[WIP] fix pooling

[rkurchin]

Goals:

• fix edge case issue of calculating pool parameters where pad can end up larger than dim and give guaranteed zeros at the edges (was causing NaN errors in training, plus just doesn't make sense)
• merge max and mean pool into one pool type to avoid duplicate code
• new type will precalculate and store pool parameters to speed up forward pass

Currently, having some issues with the that last bullet point involving the PoolDims object which I'm corresponding with @DhairyaLGandhi on Slack about.

Still to do:

• fix aforementioned issue
• finish updating tests to use new type

@paulxshen, tagged you as reviewer. I know you're probably still familiarizing yourself with the codebase, so take your time, please ask for any clarification you need, and no need to look at this until Dhairya and I sort out the issue I'm having. Meantime, I'll keep working on docs :)

[DhairyaLGandhi]

Is amortizing the cost of computing the pooldims significant?

[rkurchin]

To be honest, I'm not sure, I haven't done a detailed set of timings, but it seemed like it should be a straightforward thing to do and would certainly save time, even if I don't know exactly how much. If it's complicated in a way that I'm not seeing, I can obviously undo it easily, it just seemed like it should probably be an easy fix...is there something I'm missing here?

[paulxshen]

Got it Rachel - will look over. Thanks!

[codecov-io]

Codecov Report

Merging #31 (a0b2421) into master (163d2a3) will decrease coverage by 2.45%.
The diff coverage is 85.71%.

@@            Coverage Diff             @@
##           master      #31      +/-   ##
==========================================
- Coverage   75.00%   72.54%   -2.46%     
==========================================
  Files           3        3              
  Lines          56       51       -5     
==========================================
- Hits           42       37       -5     
  Misses         14       14              

Impacted Files	Coverage Δ
src/AtomicGraphNets.jl	100.00% <ø> (ø)
src/layers.jl	71.42% <84.61%> (-4.58%)	⬇️
src/models.jl	100.00% <100.00%> (+40.00%)	⬆️

---

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 163d2a3...a0b2421. Read the comment docs.