Expand quantity detection to large units without a "Net" symbol

[CloCkWeRX]

Right now, the model seems to heavily rely on the presence of a NET indicator to determine which is most the likely weight.

We should consider looking for units like "1 L" or "1 KG", which are too large to be commonly found in a nutrient table; where no other quantities have been detected.

Examples:
https://world.openfoodfacts.org/product/9310113900069

We may need to put sane limits (1-2 digits only?); to avoid accidentally matching country labelling for some units like litres.

It may be safe to look for any matches > 100g / 100ml

[raphael0202]

We currently look at units such as "200 ml". "1 L" raised too many false positive to be used.
See

robotoff/robotoff/insights/ocr/product_weight.py

Line 76 in e5534b4

'no_mention': OCRRegex(

I'm sure the product weight detection logic can be improved though