generalize sparsity tooling to handle `dsCMatrix` (symmetric matrix) input (low priority)

[paciorek]

At present, it seems that we need the sparse matrix format that is passed from R to be dgCMatrix.

dgSMatrix is a type for symmetric matrices and can be automatically produced from Matrix-based manipulations:

Q <- matrix(0, 4, 4)
Q[1,2] <- Q[2,1] <- Q[2,3] <- Q[3,2] <- Q[3,4] <- Q[4,3] <- -1
diag(Q) <- c(1,2,2,1)
Q <- Matrix(Q, sparse = TRUE)
class(Q)
## [1] "dsCMatrix"

myfun <- nFunction(
    fun = function(x = 'numericVector', Q = 'nSparseMatrix') {
           loglik <- sum(x * (Q%*%x))
        return(loglik)
    }, returnType = 'numericScalar')
cmyfun <- nCompile(myfun)
cmyfun(x, Q) 
# Error: Need S4 class dgCMatrix for a sparse matrix
cmyfun(x, as(Q, 'dgCMatrix')) # works

Given the ease of converting to dgCMatrix and the helpful error message, this isn't a priority, but it would be nice to allow symmetric matrices at some point for user flexibility and presumably a bit of additional computational efficiency. There might also be cases where what Eigen does under the hood in terms of algorithms matters if symmetry is known.