fix: blockitens with multiple models

[buuhuu]

• [x] parse block items with all applicable models, select the one that extracted most properties
• [x] improved detection of images and links in element groups and outside take into account how many properties a semantic field (w/ collapsing) extracts and take the one that extracts most properties.
• [x] tests