Do more to help with analyzing local/non-Cumulus data (aggregate mode)

[mikix]

Outside of a Cumulus context (i.e. in aggregate output mode), data-metrics has some feature gaps. This is a ticket to unify discussion of that (and probably actual work will be documented in other tickets and PRs). This is also related to the presentation-focused #35.

Some thoughts on improvements:

• In order to handle real hospital scale data, we need to improve the memory-usage story in ndjson mode. We can't simply keep loading it all into memory. (should be done on cumulus-library side)
• Support reading from a non-Cumulus ndjson directory layout. (i.e. instead of looking at dir/patient/*.ndjson, consider also looking at dir/*.Patient.*.ndjson) (Support a flat ndjson dir layout as well as a hierarchical one cumulus-library#244)
• Support looking at data that we don't have in Cumulus land (Consider supporting a non-cumulus-quirks mode #23 in general and DocumentReference attachment/url check #7 for a DocRef specific one)