Batch pretranslate on previously built Nmt Engine

[johnml1135]

This would involve a few parts to implement seamlessly, but in general I would imagine this user scenario:

• The user built an engine and is drafting
• Low and behold, they want a different book drafted than they originally thought they wanted! They request a "fast draft" of that book.
• 5 minutes later, they have the book drafted!

The API workflow could look like this:

• The Nmt engine is marked for persistence.
• The user requests another build and either:
  • Has the same exact training corpora? Has no training corpora?
  • Has a special flag in the options?
• it would also be helpful to have queue name so that we can assign it to a fast turn GPU to get a nice 5-20 minute turn, rather than 10 hours - New NMT option - choose ClearML queue #465

[johnml1135]

Two potential ways to invoke:

• If everything is the same (same corpora, same revisions of files, same build params)
• If explicitly invoked

[johnml1135]

Duplicate of #350.