Improvements to API for sparse matrices

[rileyjmurray]

Everything in the plan below is implemented and tested.

Plan

• make sparse matrix arguments const unless mutation is clearly necessary.
  • sketching functions
  • spmm dispatch functions
  • coo spmm kernel (rather, dispatcher to CSC kernels).
  • coo_to_cs[r/c] conversion functions.
• dispatch efficient matrix-multiply for sparse sketching with row-major data (will resolve Sparse sketching on row-major data dispatches inefficient kernel #131).
• Allocation of COO sparse matrix data should zero-initialize all arrays. (resolves reserve_coo should zero-initialize all memory #125)
• utility function for applying a symmetric permutation to a COOMatrix.
• add deepcopy() instance methods and free functions.
  • CSCMatrix
  • CSRMatrix
  • COOMatrix
• transpose instance methods for sparse matrix classes.
  • CSCMatrix
  • CSRMatrix
  • COOMatrix
• eliminate allocation of temporary vals arrays in spmm kernels when alpha != 0.
• format conversion instance methods
  • COO --> CSC or CSR.
  • CSR --> COO
  • CSC --> COO
• COOMatrix instance method for sorting

other changes

• SparseDist and DenseDist
  • have constructors that accept any ordinals for dimensional arguments, rather than only int64_t. We round down floats as needed.
  • add sample(RNGState &seed_state) instance functions.

[rileyjmurray]

EDIT: the issue below has been resolved.

---

There's web doc example text that includes the following lines

SkDist D1( d,     m, arg3, Axis::Short );
SkDist D(  d, m + k, arg3, Axis::Short );
// Since d < m and we're short-axis major, the columns of matrices sampled from
// D1 or D1 will be sampled i.i.d. from some distribution on R^d.
using SkOp = typename SkDist::distribution_t;
SkOp S1( D1,  seed_state ); // seed_state is some RNGState.
SkOp  S(  D,  seed_state );
// With these definitions, S1 is *always* equal to the first m columns of S.
// We recover S2 by working implicitly with the trailing k columns of S.

The typedef of SkOp is going backward. Right now we can't programmatically get a handle on the SketchingOperator class associated to a SketchingDistribution.