Faster implementation of cosine similarity when the vectors are fixed in the vocabulary and vocabulary is limited

[LysSanzMoreta]

When the vector per sequence element is fixed, one letter = one vector, and the vocabulary is limited (i.e amino acids) . Then the cosine similarity can be pre-computed and avoid some computations.

Tasks:

• Adapt current implementation to intake pre-computed cosine similarities

[LysSanzMoreta]

not needed now