New interpreter: Self contained Python Data Loaders with uv run scripts

[metafeather]

A suggestion to add a new default interpreter to observablehq.config.js:

    interpreters: {
        ".uv.py": ["uv", "run", "--quiet", "--no-project"]
    }

This allows the use of Python self-contained UV scripts that define inline dependencies and avoid the need for a .venv to be compatible across Data Loaders, e.g.

#!/usr/bin/env -S uv run --quiet --no-project
# /// script
# requires-python = ">=3.11"
# dependencies = [
#     "pandas",
#     "pyarrow",
#     "fastparquet",
# ]
# ///

# Outputs dataset to stdout for use with Observable Framework:
#   ./src/data/taxis.parquet.uv.py
# 
# Requires extension registered in observablehq.config.js:
#     interpreters: {
#         ".uv.py": ["uv", "run", "--quiet", "--no-project"]
#     }

import pandas as pd
import sys
import io

# URL for a recent Yellow Taxi Trip Data (January 2023 is reliably available on public S3)
DATA_URL = "https://d37ci6vzurychx.cloudfront.net/trip-data/yellow_tripdata_2023-01.parquet"

def main():
    try:
        # Define columns to keep the dataset light
        columns = [
            "tpep_pickup_datetime", 
            "tpep_dropoff_datetime", 
            "passenger_count", 
            "trip_distance", 
            "PULocationID", 
            "DOLocationID", 
            "payment_type", 
            "fare_amount", 
            "tip_amount", 
            "total_amount"
        ]
        
        # Log to stderr (visible in build logs)
        sys.stderr.write(f"Reading from {DATA_URL}...\n")
        
        # Read the remote parquet file
        df = pd.read_parquet(DATA_URL, columns=columns)
        
        # Sample the data (50k rows)
        df_sample = df.sample(n=50000, random_state=42)
        sys.stderr.write(f"Sampled {len(df_sample)} rows.\n")
        
        # Write to memory buffer first to ensure valid parquet output
        # This prevents partial writes to stdout if something fails mid-stream
        buf = io.BytesIO()
        df_sample.to_parquet(buf, engine="pyarrow")
        buf.seek(0)
        
        # Flush the buffer to stdout
        sys.stdout.buffer.write(buf.getvalue())
        sys.stdout.buffer.flush()
        
        sys.stderr.write("Done.\n")
        
    except Exception as e:
        sys.stderr.write(f"Error processing data: {e}\n")
        sys.exit(1)

if __name__ == "__main__":
    main()

(Also suggested for Notebook Kit: observablehq/notebook-kit#110)