[maint] Dependency on scipy

[alexfikl]

That's a pretty big dependency, but most people doing numerics in python will likely have it installed.

Currently used in (probably more as time goes on):

• Evaluating Jacobi polynomials: from scipy.special import jacobi, eval_jacobi, roots_jacobi.
  • quad: Can probably get away with numpy.polynomial for Legendre and Chebyshev.
  • diff: Need to find how to simplify for Legendre and Chebyshev.
• Mittag-Leffler: from scipy.special import gamma, erfc and from scipy.integrate import quad.
  • Functions available in import math. Would need to vectorize for numpy?
  • quad: fixed order? Would need to ensure high precision, but probably better for performance anyway.