Re-examine churn metric

We count churn based on subscription status. We have also tried switching to using subscription/customer min&max dates as the subscription start&end and counting churn as the month after the subscription end date. But doing it in streamlit made the charts seem really off.

I think we should address this at the data model layer. This would involve adding logic to standardize the `subscription_status` field since I think different platforms may have different values for it. Maybe something like `active, on-hold, cancelled, inactive`. We could make sure cancelled is only valid for the month after a subscription ends. And when counting # churns in a given month we would just look where status = `cancelled`.

This is because currently we are counting churn as where status = `inactive`. This may be defined differently per platform, also a subscription could potentially persist through several months as `inactive` but we don't want to wrongly count it.