Reduce memory usage of the block tally step

[GertjanBisschop]

Concatenating all blocks into a huge array prior to counting can lead to excessive memory usage, see #132.
Potential solutions (not mutually exclusive):

1. Specify correct dtype instead of always using 64-bit integers.
2. Use for loops over windows while tallying. This would allow us to skip having to allocate memory for the concatenated array entirely. Numba could be used to speed up the for loop if speed becomes an issue because of it.