Update notebooks to derive summary statistics

[LisaHopcroft]

This PR contains changes necessary to derive summary statistics of the opioid prescribing reductions identified by the change detection methodology (analysis continues to use the cached dataset).

This includes:

• changes to lib/outliers.py to capture all changes before filtering and returning them from the filtered_sparkline() function
• additions to the notebooks to filter those changes for reductions (i.e., where the level is >0) and then calculate/present the summary statistics

[LisaHopcroft]

I'm getting a warning when doing the final multiplication by 100 when generating the summary statistics table. Interested to hear if @alexwalkerepi would do this any differently.