Align built-in pretranslation with the pretranslation upload API

POST /api/v2/upload/pretranslations/ (#4511) makes different choices from the built-in pretranslation (pontoon/pretranslation/tasks.py). We agreed to bring the two in line rather than let them drift.

Strings with unreviewed suggestions

The built-in pretranslation skips any entity with a non-rejected translation. The API pretranslates it: a matching suggestion becomes the active pretranslation, keeping its author; other suggestions stay unreviewed.

Fuzzy translations already in Pontoon.

Skipped by the same filter above; the API rejects and replaces them.

Failing checks

Both run run_checks(use_tt_checks=False), so TT checks are already disabled in both and should stay disabled. But the built-in retries once with preserve_placeables=True and stores the result unchecked (tasks.py:126-135), then flags it via bulk_run_checks() and skips it during sync; the API drops it, stores nothing, and reports it in failed_checks. → Decide whether a failing pretranslation is stored and flagged, or refused.

See thread

Compare the built-in pretranslation in pontoon/pretranslation/tasks.py, especially tasks.py:126-135, with POST /api/v2/upload/pretranslations/ and the discussion in pull request #4511. First resolve how failing pretranslations should be handled; done means both paths make the same choices for unreviewed suggestions, fuzzy translations, and failing checks.