Improve labelling and filtering of match candidates (e.g. duplicates)

[nebfield]

This PR has evolved into a big refactoring of how multiple match candidates are post processed and logs are produced. Briefly:

• Move from a join-based approach to window functions to guarantee no rows can go missing
• Keep track of variants for each accession in scoring files using row_nr
• Refactor postprocess into:
  • label: Add annotations to match candidates
  • filter: Filter score accessions based on overlap % and variants based on flags
  • log: Make a summary log and a big log from raw labelled and filtered labelled data (and compress!)
• Handle and label edge cases when labelling matches in a more transparent way