[Backfill corrections] Align daily and rollup file formats; make dates portable

[nmdefries]

Description

Make sure lag and issue_date fields are included in both daily and combined files. Store dates and location info as strings for better portability. These were previously datetime64 (causing the timezone issue) and object types.

claims_hosp backfill file generation was never merged, so changes to those functions are not included here.

Changelog

• quidel_covidtest's backfill.py
• changehc's backfill.py

Fixes

The original time_values are meant to be plain dates ("2020-01-01") with no timestamp or timezone info.

The parquet format uses a schema with types. So if Python writes the time_values as either pure dates (no timestamp info) or strings, R will read them in using the same types. The implicit timezone conversion (from no timezone, which R's arrow::read_parquet interprets as UTC, to the local timezone) only happens when the time_values are saved into parquet form as datetime64s.

Python doesn't seem to have a "pure" date class (no time/timezone info). The datetime64 type assumes the time is 00:00 even if none is given. To drop the time info, we can convert to a pure date but this changes the type to object. The object class can be a little dangerous to use. In this case, it's not clear what type parquet will assign to such a column or how R will read it in, and within Python objects can behave in unusual ways. Saving the dates to string is safer.

R will have to do an extra step to convert from string to a date, but it should avoid any weird time/timezone issues.

[jingjtang]

LGTM

[nmdefries]

@krivard This is ready to merge