Fix algorithm to unpublish a document

The current algorithm is doing the following steps:

1. run a query to get all the children of a given section (without the hidden docs and the publish spaces)
2. all the DocumentModels which match the previous query are loaded into memory : if a section contains 10 thousands (or more) of published documents in the section, it may consume all of the available heap with 3 or 4 concurrent threads.
3. it loops over all (published) documents to find which ones are a published version of a given document
4. then the matching documents are deleted/unpublished

Iterating over all the children of a section could be avoided by leveraging an existing property, ecm:proxyVersionableId, which stores the id of the source document which was published to this section.

Then a query like

SELECT * FROM Document WHERE ecm:parentId = 'sectionId' AND ecm:isProxy = 1 AND ecm:proxyVersionableId = 'documentId'

would be more efficient and load only the required documentModel into memory.

NB: not sure which property is the best between proxyVersionableId and versionVersionableId