[ENH] Adapting `AptaNet` and `AptaTrans` for `Benchmarking`

[satvshr]

AptaNet and AptaTrans need a unified input format so that they both can be used for benchmarking. The idea is to follow what is said in #106.

[satvshr]

I had a question @fkiraly @NennoMP :
I thought AptaTransPipeline was supposed to have fit and predict methods following sklearn api but it does not, so how is benchmarking supposed to run if there is no common method you can call from both (the other being the aptanetpipeline) classes in order to fit and get predictions?

Just a few generic statements:

1. prot_words description needs to be changed everywhere it is being used with a genric description that will be copy pasted everywhere, its way too confusing.
2. Smaller comment, predict_api should mention the datatype being returned in the Tensor, currently I have to look into evaluate in order to understand what the function is returning.