[FEA] Exploit Symmetry for `groupby.corr` and `groupby.cov`

In https://github.com/rapidsai/cudf/pull/9889#discussion_r771493338, @bdice made an observation that due to symmetry, there's a chance of utilizing that to reduce computation by only applying the upper half of the matrix. In an attempt to reduce the scope of the PR, this is the issue that tracks the suggestion separately. If applicable, `groupby.corr` should also share the logic.

cc @skirui-source