Fix ParameterlessLayer conformances.

[dan-zheng]

Associated type inference behavior was changed in swiftlang/swift#32578:
derived conformances are now attempted before associated type inference.

This broke ParameterlessLayer, which relied on a
TangentVector == EmptyTangentVector same-type constraint to set a
default TangentVector type witness for conforming types.

Add explicit TangentVector type witnesses to
ParameterlessLayer-conforming types to fix this regression.

This workaround is forward- and backward-compatible, but makes code more
verbose.

Related tensorflow/swift-apis change: tensorflow/swift-apis#1034

---

Fixes build errors from the 2020-06-30 master -> tensorflow merge:

/swift-models/Models/ImageClassification/ShuffleNetV2.swift:21:15: error: 'ParameterlessLayer' requires the types 'ChannelShuffle.TangentVector' and 'EmptyTangentVector' be equivalent
public struct ChannelShuffle: ParameterlessLayer {
              ^
/swift-models/Models/ImageClassification/ShuffleNetV2.swift:21:15: note: requirement specified as 'Self.TangentVector' == 'EmptyTangentVector' [with Self = ChannelShuffle]
public struct ChannelShuffle: ParameterlessLayer {
              ^
/swift-models/Models/ImageClassification/ShuffleNetV2.swift:21:15: error: type 'ChannelShuffle.TangentVector' does not conform to protocol 'VectorProtocol'
public struct ChannelShuffle: ParameterlessLayer {
              ^
[19/147] Compiling TextModels BERTClassifier.swift
/swift-models/FastStyleTransfer/Layers/Helpers.swift:4:15: error: 'ParameterlessLayer' requires the types 'ReflectionPad2D<Scalar>.TangentVector' and 'EmptyTangentVector' be equivalent
public struct ReflectionPad2D<Scalar: TensorFlowFloatingPoint>: ParameterlessLayer {
              ^
/swift-models/FastStyleTransfer/Layers/Helpers.swift:4:15: note: requirement specified as 'Self.TangentVector' == 'EmptyTangentVector' [with Self = ReflectionPad2D<Scalar>]
public struct ReflectionPad2D<Scalar: TensorFlowFloatingPoint>: ParameterlessLayer {
              ^
/swift-models/FastStyleTransfer/Layers/Helpers.swift:4:15: error: type 'ReflectionPad2D<Scalar>.TangentVector' does not conform to protocol 'VectorProtocol'
public struct ReflectionPad2D<Scalar: TensorFlowFloatingPoint>: ParameterlessLayer {
              ^
/swift-models/FastStyleTransfer/Layers/Helpers.swift:41:15: error: 'ParameterlessLayer' requires the types 'ReLU<Scalar>.TangentVector' and 'EmptyTangentVector' be equivalent
public struct ReLU<Scalar: TensorFlowFloatingPoint>: ParameterlessLayer {
              ^
/swift-models/FastStyleTransfer/Layers/Helpers.swift:41:15: note: requirement specified as 'Self.TangentVector' == 'EmptyTangentVector' [with Self = ReLU<Scalar>]
public struct ReLU<Scalar: TensorFlowFloatingPoint>: ParameterlessLayer {
              ^
/swift-models/FastStyleTransfer/Layers/Helpers.swift:41:15: error: type 'ReLU<Scalar>.TangentVector' does not conform to protocol 'VectorProtocol'
public struct ReLU<Scalar: TensorFlowFloatingPoint>: ParameterlessLayer {
              ^
/swift-models/FastStyleTransfer/Models/TransformerNet.swift:5:15: error: type 'TransformerNet.TangentVector' does not conform to protocol 'VectorProtocol'
public struct TransformerNet: Layer {
              ^
/swift-models/FastStyleTransfer/Models/TransformerNet.swift:58:15: error: type 'ConvLayer.TangentVector' does not conform to protocol 'VectorProtocol'
public struct ConvLayer: Layer {
              ^
/swift-models/FastStyleTransfer/Models/TransformerNet.swift:90:15: error: type 'ResidualBlock.TangentVector' does not conform to protocol 'VectorProtocol'
public struct ResidualBlock: Layer {
              ^
/swift-models/FastStyleTransfer/Models/TransformerNet.swift:124:15: error: type 'UpsampleConvLayer.TangentVector' does not conform to protocol 'VectorProtocol'
public struct UpsampleConvLayer: Layer {
              ^