add mobilenet v3 (#409)

* add mobilenet v3

* fix filter shape mismatch

* add some tests + lint

* apply filter mult to input conv as well, unify styles

* swap hard sigmoid back in

* add hard sigmoid to initial block as well

* add zeropad before initial block

* add dropout

* move stride up a layer

* initial dropout rate tweak

* refactor activation bool to enum type

* refactor to use seblock layer

* swift-format pass

* fix se block conv type

Author: brettkoonce

Models/ImageClassification/CMakeLists.txt

@@ -3,6 +3,7 @@ add_library(ImageClassificationModels
   LeNet-5.swift
   MobileNetV1.swift
   MobileNetV2.swift
+  MobileNetV3.swift
   ResNet.swift
   ResNetV2.swift
   SqueezeNet.swift