Implement a method for specifying row/col sites

This allows a user to specify an array of row sites to be compared
against an array of column sites. Rectangular matricies are returned as
a result of these comparisons. We weighed the pros/cons of the
indexing performance and redundant computation. The initial
implementation avoided redundant computation, but at the expense of
complexity and readability. The final implementation is much more simple
and DOES perform redundant computations if the row/col sites cross the
diagonal of the LD matrix. We will revisit if/when this becomes a
problem.

In addition to these changes, I've cleaned up the results matrix
allocation. Since the python code will be allocating and managing this
memory, we'll pull out the allocation code.

C tests have been added to capture all of this new functionality.

A type annotated python prototype that mirrors the C functionality was
added to provide some documentation on how the C algorithms work, and
to aid in the testing of the C code once the python API is ready.