Merge pull request FluxML#362 from FluxML/cuda-1.0

1.0 CUDA support

Merge pull request FluxML#356 from domluna/recurrent-fix

recurrent bug fixes

Merge pull request FluxML#351 from FluxML/fbot/deps

Fix deprecations

add statsbase to require

perf hacks

gradient hook

Merge pull request FluxML#255 from FluxML/ad-docs

AD Docs

closes FluxML#198

bson api

load data