called prior to transformations now by default

[Test] Add test for QuantRelu: per_channel_broadcastable_shape is not None but scaling_per_channel is False

Scaled and non-scaled QuantHardTanh activation tests separated. Non-scaled test includes cases with scale parameter == 1. In scaled test, bipolar activation is excluded as it's not currently supported

Add new test cases for QuantHardTanh. Cases for max > 1 and scaling_impl_type == ScalingImplType.PARAMETER with learned scale

thanks to @quetric for pointing this out