add index_bits option to support large datasets

- add `index_bits` optional argument to KDTree
  - default value is 32: preserves existing behaviour & performance
  - user can specify 64 instead to use 64-bit integers
    - this ensures correct results when n_points * k > 2^32
    - uses approx 50% more RAM than 32 bit option
    - resolves storpipfugl#38
- in 32-bit int mode add checks to avoid returning incorrect results
  - KDTree checks that `n_points < 2^32`
  - KDTree.query checks that `n_points * k < 2^32`
- update tests
  - parametrize all existing tests to test 32-bit and 64-bit int mode
  - add a query test with n_points * k too large
  - didn't add a test with n_points too large as it would require 16GB RAM to run