Supporting ADBC "numeric" data

[diviyank]

Description

Hello everyone,

As of today, the ADBC driver for PostgreSQL will output all PostgreSQL NUMERIC columns as Arrow strings (see apache/arrow-adbc#767)

And polars does not support that type yet:

line 1213, in arrow_to_pydf
    pydf = PyDataFrame.from_arrow_record_batches(tbl.to_batches())
           │           │                         │   └ <cyfunction Table.to_batches at 0x7f922ffc35e0>
           │           │                         └ pyarrow.Table
           │           │                           assortment_id: int64
           │           │                           sale_start_date: timestamp[us, tz=UTC]
           │           │                           sale_end_date: timestamp[us, tz=UTC]
           │           │                           canal_id: int6...
           │           └ <staticmethod(<built-in method from_arrow_record_batches of type object at 0x558e4899bde0>)>
           └ <class 'builtins.PyDataFrame'>

polars.exceptions.ComputeError: cannot create series from Extension("arrow.opaque", Utf8, Some("{\"type_name\":\"numeric\",\"vendor_name\":\"PostgreSQL\"}"))

This occurs when performing operations on BIGINT columns in the database, that immediately produces NUMERIC types columns.

One way to circumvent the issue at the moment is to manually cast the output type of the columns in the query as the desired type : BIGINT or STR or something else.

Would it be possible to enable conversion for this type?

Thank you!