50% of work is copying from C++ to Python when there are many matches

[unzvfu]

The code

        for j in range(matches):
            ind = c_indices[j]
            assert ind < len(filters2)
            result.append((i, c_scores[j], ind))

at entitymatcher.py:93 is responsible for over 50% of the work when there are many matches (e.g. large k ~ len(filters2) and small(ish) threshold ~ 0.5).

This could be (partly) resolved by a solution to issue #67. (Issue #66 is also relevant.)

[unzvfu]

Fixed in #101.