Recognition find batched

[kremnik]

Tickets

#1332

What has been done

With this PR, the find_batched method was developed as a batched version of the find method. The main difference of that method is using numpy arrays instead of pd.DataFrame and using batched distance calculation instead of element-wise distance calculation. Also, the function returns array of array of dicts (where array[:] is source faces and array[:][:] is founded faces corresponding to each source face) instead of array of DataFrames.

Also, the tests for the find_batched method was developed in the tests/test_find_batched.py.

Based on the experiments conducted on large face databases, which include photos with multiple faces, the find_batched method achieves a speedup of more than 50% compared to the find method.

How to test

make lint && make test

[serengil]

this is a comprehensive PR, need to spend some time on it

[serengil]

as a future PR, we may consider to move find_distance_batch, find_euclidean_distance_batch and find_cosine_distance_batch to the functions in verification.

[serengil]

LGTM