feat: initial completeness-tracking work

[mikix]

This is early support for completeness-tracking (the ability to mark which groups & resources have been loaded by the ETL and are thus ready for studies to use).

• Adds some new (secret) CLI arguments: --export-group, --export-timestamp, and --write-completion
• Adds a new etl__completion table which holds:
  • table
  • group
  • export_time
• Adds a new etl__completion_encounters table which holds:
  • encounter_id
  • group
  • export_time
• This table is automatically written to, using the CLI values
• Currently, those arguments are optional. A future change will make them required. (though hopefully usually automatically inferred from export logs)
• When using the ndjson output format, you can no longer have any files in the output folder. This is to safeguard against accidents (and to make some code paths simpler)

This PR starts to address #296 (Library side here)

Future Plans
This PR is just an preparatory, disabled-by-default step 1. These are bits of work that still have to be done to make this feature real (and required) across Cumulus.

• Parse export logs to autodetect arg values
• Require the new ETL args (from user, from log, or from our own export)
• Prevent overwriting new group data with old, to solve Epic use cases (check dates before deltalake write)
• Generate export logs from our own bulk export implementation, allowing us to parse it back later
• Backfill/migration support (maybe as simple as running --group-name legacy --timestamp 2020 for all current encounters)
• Handle empty input sets - may require patching the bulk exporter to leave an empty file in those cases
• User docs

Checklist

• Consider if documentation (like in docs/) needs to be updated
• Consider if tests should be added