Running GSVA without kernel density estimation on a dgCMatrix coerces to a dense matrix

[axelklenk]

explore if we can estimate ECDF using a sparse matrix directly, i.e., without coercing to a dense representation; start by using the pure R implementation of ECDF