feat: add MS1 scan extraction to parquet conversion (--include-ms1)

[BioGeek]

Summary

Extend mzML/mzXML readers and the instanovo convert CLI to extract MS1 scans alongside MS2 scans into the same Parquet files.

New --include-ms1 flag

# MS2 only (default, backward compatible)
instanovo convert data.mzML output/ --name sample --partition test

# MS1 + MS2 in same Parquet files
instanovo convert data.mzML output/ --name sample --partition test --include-ms1

Schema changes

New ms_level column added to the data dict and resulting Parquet:

• 1 for MS1 scans
• 2 for MS2 scans

MS1 rows have None for precursor_mz, precursor_charge, and sequence (these fields are not applicable for MS1 spectra).

Files changed

• instanovo/utils/msreader.py: read_mzml() and read_mzxml() accept ms_levels parameter (default [2] for backward compat). read_mgf() adds ms_level=2 for consistency.
• instanovo/utils/data_handler.py: load_mzml() and _df_from_mzml() pass through ms_levels.
• instanovo/scripts/convert_to_sdf.py: New --include-ms1 CLI flag. When set, loads MS1+MS2 from mzML/mzXML; warns and ignores for other formats.

Motivation

The INFlow pipeline uses Parquet as the sole data substrate after conversion. The meeting decision (2026-04-10) settled that:

• MS1 and MS2 scans go in the same Parquet files with an ms_level column
• INQuant needs MS1 data for quantification
• Winnow-related metrics may also use MS1 signal information
• Downstream processes filter by ms_level as needed (e.g., InstaNovo predict uses only ms_level=2)

Test plan

• test_convert_command passes
• test_mzml_to_parquet passes
• Pre-commit hooks pass (ruff, mypy, codespell)
• Backward compatible: default behavior (MS2 only) unchanged

🤖 Generated with Claude Code

[BioGeek]

Additional fix added to this PR

Always include experiment_name and spectrum_id in mzML/mzXML output (988c1ca)

All three readers (read_mzml, read_mzxml, read_mgf) now add experiment_name (from filename stem) to each spectrum row. _df_from_dict() auto-constructs spectrum_id as experiment_name:scan_number.

This ensures Parquet output from instanovo convert always has the columns needed by Winnow (which merges on spectrum_id). Previously these columns were only added through the SpectrumDataFrame constructor path with add_spectrum_id=True, not the direct load_mzml() path.

All tests pass.