Refine numeric column handling

[hrecht]

Detecting numeric columns is messy: the APIs return all data as strings. getCensus currently types columns with numbers in the col name as numeric. This works well for the ACS and decennial Census APIs but needs improvement for several reasons:

• a few string variables contain numbers in their name (e.g. fage4 in the timeseries/bds/firms API) so the current approach coerces them to NA
• some numeric variables have letter-only variable names (e.g. in the the SAHIE API)

Theoretically, it should be possible to use the variable API endpoints to get type information (the approach of Python lib census https://github.com/datamade/census) but that field doesn't uniformly exist and is often wrong. I've raised this issue with the Census dev team and hope they'll correctly type data in the future.

Until then, this package needs typing improvement particularly for the timeseries APIs.

[hrecht]

This package currently uses a set of rule-based workarounds to type data as string or numeric. Until properly formatted JSON is returned by the Census APIs (which I have asked about multiple times), this is an issue to be solved by the Census Bureau.