fix edge cases during semver grouping creating single PRs erroneously

[jakecoffman]

This pushes the "handled" logic into the group so that the compile_updates_for method can specify that the dependency is handled immediately when it is found to be in the semver grouping. This way, if there is an error or the update is not possible, it will still be considered handled and not attempt to create a single PR for it.

Also, if an exception is thrown, even during version fetching before we can know whether it's in the grouping or not, we consider the dependency handled as well since it's not known what versions are available. Customers can request a rebase or recreate to try again and not have superfluous PRs kicking about.

[jakecoffman]

The specs have shown an issue and now I'm seeing both groups and dependency_snapshots have dependencies. It seems like the snapshot has the dependencies that are already filtered on by patterns, and the group dependencies are populated as the update proceeds? A little unclear...

Wait, no the dependency engine populates the group dependencies... !?

Edit 74: After studying the code, I think the failing specs are misleading. Now that we're in the world where ungrouped_dependencies is only populated by actually doing an update, they only make sense if add_to_handle is called. So I updated the test.

[jakecoffman]

I need to add a test and then this will be ready to go.