I think without a free starting point of some kind, this is going to be a tough sell. Perhaps a week free or something like that would enable someone to see if they like it enough to start paying for it.
Great, it'd be sweet to have you involved! FFT, polynomial fit, and d3.js would be huge steps. For me, the most important thing is ensuring that all changes are application driven. Building a library that's designed to be used is top priority.
Sparse and dense matrix math and eigensolvers would be really nice. Especially solvers that let you efficiently get the eigenvectors associated with the smallest eigenvalues as this comes up in graph and image partitioning.