Improve quantity exploration and cleaning notebooks

[TheSanjBot]

• explores product_quantity_unit, product_quantity, and quantity
• exploration to understand variation and missingness
• cleaning logic for OCR cleanup, unit normalization, multipack parsing, conflict handling
• keeps true contradictions as conflicts and ambiguous cases unresolved
• helps prepare quantity data for more reliable downstream product consolidation

The outputs:

• resolved
  • used when the quantity can be interpreted confidently in a comparable normalized form
  • examples: 500 g, 12 x 355 ml, 24 g (0.85 oz)
• conflict
  • used when quantity signals materially disagree, so the pipeline should not guess
  • example: 25 kg / 5 lbs
• partial
  • used when the row contains useful quantity-related information, but not a safely comparable package size
  • examples: packaging-only information
• unresolved
  • used when the text remains ambiguous, broken, or unsupported

The current workflow is useful, but not complete:

• many rows are still unresolved because the logic is intentionally conservative
• some mixed expressions still need attention
• multilingual alias coverage is helpful but not comprehensive of all languages
• some rows are only rescued because structured fields are reliable, not because the text parser fully understands them
• household-unit expressions are intentionally excluded from consolidation
• a small set of remaining conflicts may still be recoverable with additional safe rules