When scanning structured documents, use the structure!

We are scanning markup as plain text such as a JSON, YAML, XML, or HTML files and so on (not when these are package manifests).
We should instead parse these are treat them as a structure. For instance, it does not make sense to detect a single license across multiple attributes in a JSON file.

This would help removing a class of false positive. We are already doing special pre-processing for PDF and binaries... This would just extend the practice to more file formats.