[Prototype] Client-side caching option for Epidata.covidcast Python

[dshemetov]

Description

Epidata.covidcast is idempotent on the hour-scale (I think). This PR adds client-side caching to Epidata.covidcast in Python, to save user effort in writing their own file saving mechanisms.

Example:

# works as normal
> df = Epidata.covidcast('source', 'signal', 'day', 'county', 20200414, '01234')
# a regular call to the API that caches locally
> df = Epidata.covidcast('source', 'signal', 'day', 'county', 20200414, '01234', cache_timeout=20)
# performs a local lookup (if run within 20s)
> df = Epidata.covidcast('source', 'signal', 'day', 'county', 20200414, '01234', cache_timeout=20)

The goal is to expose cache_timeout to the covidcast Python library in another PR. We can keep the default behavior cacheless to remove hidden state issues.

Overlapping, but not identical time windows will look to the cache and the web gracefully (because the covidcast Python library splits date range requests into daily Epidata.covidcast calls). This is not true for geos (but could be if we can provide metadata of available geos per indicator).

Changes

Switch to requests_cache when the cache_timeout argument is sent.

[chinandrew]

nit: check that the length of the call args (number of calls) is correct. otherwise lgtm

[krivard]

Are any of these changes:

• breaking
• announce-able
• needing public documentation