Implement covariance matrix as a sparse matrix

[phillipstanleymarbell]

Related to #128 and #119.

[harrysarson]

I am working on some RISC-V fixes and cannot verify the patches because sunflower crashes when trying to allocate uncertain memory.

The implementation of a sparse covariance matrix could be done as follows:

• Have a dense covariance matrix of size N << mem_size.
• Use a hashmap to convert memory addresses to rows in the dense covariance matrix.
• If the dense covariance matrix is "full" and the program trys to store an uncertain-float in at a new memory address simply crash the program.

I would be happy to have a go at implementing this (if only because I could then verify my other patches). The question is how to get a hash map implementation in sunflower? @phillipstanleymarbell could I compile in a small bit of c++? Is there a better way?