Parallel iterators should be supported on uninit-based matrices

Currently `par_column_iter_mut` (and its non-mutable variant, although that one is less relevant here) is only implemented for `T: Scalar`, which makes it impossible to use on matrices created via `DMatrix::uninit`.

I don't think this bound is necessary, and it makes it impossible to safely fill out an uninitialized matrix in parallel before marking it as initialized.