[feat] MMFT MLP head output key; mix model losses and MMFLosses

[apsdehal]

This will allow doing multitasking on single batch easily through different heads with different output_key per task

Mixing allows usage of heads that returns losses directly with heads that don't and rely on MMFLoss

Test Plan: Tests have been added

[facebook-github-bot]

@apsdehal has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[facebook-github-bot]

@apsdehal has updated the pull request. You must reimport the pull request before landing.

[facebook-github-bot]

@apsdehal has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.