add optional 'touched packages' metric to resolves

It would be useful to know what packages were considered during a resolve. This should help identify issues where iterations back to old versions causes slow resolves

No file or test is named. Start by locating the resolve implementation and any existing metrics, then determine how an optional touched-packages metric should expose packages considered during a resolve. Done means the metric reports the considered packages without changing default resolve behavior, with coverage for the reported result.

Written by the indexing model from the issue text.