Provide way to avoid OOM with county-level data access

[brookslogan]

This query

dat <- epidatr::pub_covidcast("doctor-visits", "smoothed_adj_cli", "county", "day")

seems to produce a nondeterministic mix of problems:

• "Operation timed out after 900006 milliseconds with 0 bytes received" (on @XuedaShen's system)
• Raising an error because of low bytes/sec (on my system)
• (Potentially making the system unresponsive and then) getting killed by out-of-memory (OOM) killer (on my system). I'm estimating (from state data * rough # counties per state) that data as an object would be ~0.7GB, and as a JSON string is around ~1.5GB; my system as ~32GB RAM, with 8.5GB "free" at the moment, so hypothetically it should be able to handle it in RAM without OOM killing, but disk might be an option too.

The last one seems like something epidatr could potentially help out with.