Fix COPY data types issues

[Jesus89]

Fixes #784.

This PR contains important fixes related to uploading and downloading data types 🎉

• Fix upload types: the old implementation it was iterating by the row. To get a row, panda unifies the type of all the columns in the row, so, if for example, you have a column with int and another with float it assumes that the row it type float so it will add your int column data as float causing a non-deterministic bug. Now the iteration is made using the index of the row, so the final type of the value is the correct one.
• Encode numeric types NaN, Infinity, -Infinity: tt adds a map between numpy types np.nan, np.inf and -np.inf.
• Fix dtypes to pgtypes and pgtypes to dtypes conversion: using PostgreSQL types instead: https://www.postgresql.org/docs/7.4/datatype.html. Using pgtype allows a better mapping for the DataFrame types. The PG types aliases have been added too (int4, float8, etc).
• Allow upload NULL values: in the old implementation, null values were ignored or converted to ' ' . In this PR NaN and None valued in the DataFrame are properly converted to NaN and null in the DB.
• Fix download NULL and Infinite values: it allows to download null, NaN, Infinite, -Infinite values from the CARTO table.
• Fix upload coordinates (0, 0): there is also a small fix to enable uploading the geometry POINT(0 0).

[andy-esch]

❤️Love this one! Bugs like these are maybe the most annoying! Beyond the small refactor suggestion, I don't see any recommended changes. Most of it is over my head unless I spend more time thinking through it.

[alrocar]

Some minor comments.

About tests, are we adding a e2e test to check Inifinity and NaN are properly uploaded/downloaded? Same for 0,0 coordinates.

[Jesus89]

We have just unit tests for that. I'll add a couple of e2e tests.

[alrocar]

✨

[alrocar]

Acceptance 🍏