Fix for QActivations passed as an argument

[AdrianAlan]

Fix for QKeras layers with QActivations passed as an argument.

Right now these models fail to compile with TypeError in the keras_to_hls method as the method expects the layer['activation'] to be a string, not a dict.

[thesps]

Thanks for this! Could you adapt the test to do some check on the output more than just compilation? I would suggest testing this functionality in existing test_keras_api.py, following the pattern of constructing a model, compiling, running prediction with some random inputs, checking the output Keras vs hls4ml, and doing some explicit checking that the converted model makes sense (ie here that the activation is what you expect).

[thesps]

Thanks for this! Could you adapt the test to do some check on the output more than just compilation? I would suggest testing this functionality in existing test_keras_api.py, following the pattern of constructing a model, compiling, running prediction with some random inputs, checking the output Keras vs hls4ml, and doing some explicit checking that the converted model makes sense (ie here that the activation is what you expect).

[AdrianAlan]

I moved it to test_qkeras.py, I thought it may belong there, rather than test_keras_api.py. Let me know if you have some other quantizers in mind. BTW, it solves #248