Exporting seems to convert some numbers to floats

[mikix]

Describe the bug
I noticed this when uploading data - that the version was being conveyed as a float (e.g. 1.0) in the parquet despite being defined as an int.

This only affects the parquet files, not the csv files. This may affect other code paths too, but I only happened to notice it with the version.

To Reproduce
Steps to reproduce the behavior:

1. Take a study that has a version defined like:

CREATE TABLE data_metrics__meta_version AS
SELECT 1 AS data_package_version;

(You can even add CAST(1 AS INTEGER) if you like.
3. Build and export it.
4. Inspect the parquet file - it holds a float type there.

Expected behavior
I'd expect integer type.

Additional context
This seems to be due to DuckDB's col_pyarrow_types_from_sql (called by exporter.py) converting NUMBER to float64. However, this is not wrong per se:

>>> duckdb.sql("SELECT 1 AS i")
┌───────┐
│   i   │
│ int32 │
├───────┤
│     1 │
└───────┘
>>> duckdb.sql("SELECT 1 AS i").description
[('i', 'NUMBER', None, None, None, None, None)]
>>> duckdb.sql("SELECT CAST(1 AS FLOAT) AS i").description
[('i', 'NUMBER', None, None, None, None, None)]

So... ideally we'd ask duckdb for a better schema description - this one is super vague. (Note that you can't do CAST(1 as NUMBER) -- duckdb doesn't recognize it as a type -- don't know why it's using it in this description.)