rbind, vec_rbind, bind_rows give wrong metadata

[brookslogan]

These functions should:

• check for a common other_keys else abort or do something smart
• check for a common geo_type else use "custom"
• check for a common time_type else use "custom"
• use the max of the as_ofs

[brookslogan]

Factors into #213.

[brookslogan]

Alternatively, if the as_ofs differ, maybe we should degrade into a non-epi_df object, since it looks like we aren't using a single as_of. But this might not play well with the auto-as_of tagging (epi_dfs from the same script might get different as_ofs). It might require changing the default as_of to a sentinel value like "unspecified" or special class to express "unspecified; object created at ".

[brookslogan]

A couple implementation paths:

• Go ahead and validate geo_type in dplyr_reconstruct. Worry about performance cost.
• Convert geo_values to factors or another special class when detected/specified as a particular type. Provide S3 methods that will automate degrading into another class. Have dplyr_reconstruct check for degradation.
• Previous bullet point, but just remove geo_type from the epi_df altogether; have it tracked within the geo_value column.