Add support for sampling rate in audio logging

[bartvanerp]

Resolves #149

TensorBoardLogger.jl currently writes WAVs and puts them in encoded_audio_string. That means TensorBoard will only pay attention to the sample rate stored in the WAV header. The sample rate is not written there however. It is not possible to include the header information in the FileIO.stream, hence using wavwrite and adding WAV.jl is necessary as this does support the sample rate argument.

[oxinabox]

nice work. Can we write a unit test for this?