Improvements to case-level metadata

[jnu]

A few improvements to handling of inferred data from document redaction runs.

• Use the new runtime parameters expected by the new version of bc2
• Track a new, less-structured, map of placeholders we've inferred and join them into the mask info before running redaction
• Save inferred id->mask and name->mask maps in redis with other case data, for re-use in related documents

This should all make redactions for documents within the same case more consistent. In addition, we should do a better job conforming to the subjects map sent with the request as well as producing the id->mask map in the output as promised.