Text Detection Normalization Should Include "Type"

When normalizing "/tagging/imageText" data, too much information is lost. In particular Nuxeo does not save/normalize the "type" of text that was detected.

For example when using "aws.textDetection", one of the fields that Amazon returns is "textDetections[i].type", where "type" can be "LINE" or "WORD". There is also a parent-child relationship between each word and the line it belongs too.

When Nuxeo normalizes the data, the "type" is lost so it's impossible to distinguish between a "word" or a "line" of text via the normalized metadata. This seems like an important attribute that should not be lost.