LintDiff: Move correlation logic into reconciliation

We're already incorporating data about both "before" and "after" states in the reconciliation code. Use a structure that builds out a set of expected results and then fill that structure with the results as they come in. For each "after" readme/tag match it to a "before" readme/tag. (As it is today) each autorest command for a given readme/tag combination should only be run once. 