While investigating why initial macOS attribution data had a lower attribution rate than expected I realized that one reason why this may be the case is because we're using the same file as a cache for it. This means that even if a user has paved over with a fresh install, we will pick up the previous attribution data (even if it was null).
While doing this, I think we should also accommodate having separate attribution data per channel, which should keep the data cleaner. Release, Beta, and ESR use `Firefox.app`, which makes the likelyhood of a paveover with a different channel reasonably high.
Differential Revision: https://phabricator.services.mozilla.com/D192628
2b8bc465a3