Dataframe/Numpy Array Loader

[sc1f]

This PR adds a separate load path for Pandas dataframe and numpy arrays (dictionaries of numpy arrays, numpy structured arrays, and numpy record arrays.

• It will try to copy all arrays of integral/float/bool type, and fill other types through iteration.
• Arrays of dtype=object will always be iteratively filled using the marshal method, which coerces data into the correct format.
• Support for mixed dictionaries of numpy arrays and lists has been deprecated: datasets should be fully composed of numpy.ndarray or list, but not both.
• Adds support for numpy structured arrays.
• Adds and fully tests support for all numpy datetime types from nanosecond (datetime64[ns]) to year (datetime64[Y]).
• Adds support for loading timestamps (in seconds or milliseconds since epoch) into columns already typed as datetime. To load timestamps as datetimes, either construct the table with a schema of {"name": datetime}, or with a valid datetime/Timestamp/datetime64 object.
• Adds and tests support for timedelta64 arrays, which are cast into string to preserve their units.
• Adds and tests support for pandas.DateTimeIndex and pandas.PeriodIndex.

[texodus]

Looks good!

[texodus]

Thanks for the PR!