Kernel needs for AFQMC

[meisenbach]

For the QMCPACK AFQMC Kokkos implementation, we will need support for operations on matrices of complex numbers.
Besides dense and sparse multiplies, we will need:

Matrix determinant, inversion and diagonalization.

We currently use:
LU factorization (zgetrf) and solve (zgetri / zgetrs)
LQ factorization

[crtrott]

Based on discussion we need sparse matrix / dense matrix multiply, where the dense matrix is actually just a multi vector (i.e. very skinny). Also the sparsity of the sparse matrix is not very high (0.1-0.5) and it seems like that matrix is also pretty skinny (15kx2k).

[crtrott]

Dimensions: A 1M x 10k,
Walkers ~ 16 or fewer: B 10k x ~16
A sparsity ~1-10%

Decreasing Importance:

• Sparse-Dense (actually SPMV with MultiVector) Problem defines A, and user is defining number of Walkers columns of B

[srajama1]

@meisenbach : Just confirming, you want zgetrf and the solve for batch of dense matrices, correct ?

[ndellingwood]

@srajama1 can you confirm if PR #306 by @vqd8a satisfies the requirements for this request?