use pandas Int64 type by default to avoid precision loss

[tswast]

The Int64 dtype has been around long enough now that I'd be comfortable with using it by default. In #786 I notice that float64 dtype is used for nullable integers even when this would cause loss-of-precision.

[tswast]

While we're at it, there are a couple of default dtypes we should change:

• Use boolean instead of object for nullable bool values https://pandas.pydata.org/pandas-docs/stable/user_guide/boolean.html#
• Use datetime64[ns] for date columns when within allowed range (per conventions at https://pandas.pydata.org/pandas-docs/stable/user_guide/timeseries.html#overview)

[jimfulton]

FTR, as someone who's tried to cram lots of data into pandas, int64 sounds brutal. :) It's certainly better than float.

Integer data with small range are pretty common. If I had to guess, I'd say most common.

I'm not sure what could be done. I guess some ideas:

• use limit to avoid loading too much data,
• Provide hints like, use int16 for integers by default.
• Provide an option to convert numeric with 0 scale to integers based on precision.
• Provide an option to scan data to determine ranges that inform types.
• Give up and use dask(?) or spark to work with the data.

[tswast]

They can already manually specify smaller dtypes if they'd like, which hopefully is a good middle ground. Folks that know they are working with smaller ranges can specify as such, but by default we support the full range as BigQuery does.

[tswast]

Maybe we also want to use the new "string" dtype by default if the pandas version supports it? https://pythonspeed.com/articles/pandas-string-dtype-memory/

Edit: Looking more closely at that article, it's experimental. Also it should be possible to use the existing dtypes argument to select this type if desired.