fix: Pandas parser does fail to parse integer or boolean only dataframes

[pierrecamilleri]

• fixes reading data from pandas returns null values #1678

Converting the Series returned by iterrows() to a dict converts np.int64 type to python's native int type and fixes the bug (same with booleans).

• Adding non-regression tests

• I was also concerned with the next lines, especially : if value is np.nan: value = None
  • it was untested, so I added a test. It looks like to_dict would not change the behavior of np.nan conversion (see side note), so I left this code unchanged.
• Primary keys are returned as ints or tuple[int], no np.int64 there
• Timestamps types are kept unchanged, so the if isinstance(value, pd.Timestamp): still applies.

Side note

np.nan behavior is quite strange with df.iterrows() : in a number column, it will be converted to float("nan"), whereas in string column it will be kept as np.nan. Adding to_dict() to the row Series does not change the types.

[pierrecamilleri]

@roll @pdelboca ready for review !

[pdelboca]

Looks good! I like changes followed by tests 👍🏼