Use Pandas names for Pandas datatypes

[rolyp]

The current system for naming types is overly complex: each type has an internal name (e.g. "integer") and an associated PFSM. We then separately have a map (twice – see #68) between internal names and Pandas names. This is confusing and leads to weirdness like show_schema showing the inferred type as “integer” but df.types showing Int64.

• Schema.transform suggests that date-iso-8601, date-eu and date-non-std can all be converted automatically to datatime64 by Series.astype – is this correct?
• ptype_pandas_mapping is defined more than once