Parquet input and output format support

[vmvarela]

Part of #68
Depends on #95 (format plugin architecture)

Description

Add Apache Parquet as an input and output format. This requires finding a Zig-compatible Parquet library or C bindings (e.g., Apache Arrow C Data Interface or nanoarrow).

This is a size:l issue due to the library integration complexity. A spike may be needed first to confirm feasibility.

Acceptance Criteria

• --input-format parquet reads a Parquet file from stdin or --input flag
• --output-format parquet writes results as Parquet to stdout or --output flag
• Column types are preserved (integers, floats, strings, timestamps)
• Parquet schema is inferred from query result column types
• Error message if Parquet library is not available at build time (optional feature flag)
• Tested with files generated by pandas, DuckDB, and Apache Spark

Notes

• Investigate: nanoarrow (C library, small, permissive license), parquet-go (not relevant), or building from scratch
• Parquet is columnar — reading row-by-row may be inefficient; batch reads preferred
• May want to gate this behind a compile-time feature flag (-Dparquet=true) to avoid mandatory C dependency
• Consider de-scoping to just Parquet input first (output is harder)

[github-actions]

This issue has been inactive for 30 days. It will be reviewed in the next backlog refinement. If it's still relevant, remove the stale label.