Fix exercise extra_fields migration for non-m_of_n mastery models

[rtibbles]

Summary

migrate_extra_fields was carrying over non-null m/n values for non-m_of_n mastery models (e.g. do_all, num_correct_in_a_row). The mastery criteria JSON schema requires m and n to be null for these models, causing frontend AJV validation to fail and exercises to appear incomplete — even though backend mark_complete passed them (it skips schema validation for old-style data).

Fixing just the m/n nulling in migrate_extra_fields would have been sufficient to resolve the immediate bug. However, to avoid having to worry about old-style exercise data ever again, this PR also:

• Fixes the ricecooker import path (create_node) to run migrate_extra_fields on incoming exercise data before writing to DB
• Adds a management command fix_exercise_extra_fields to repair existing data in production — both already-migrated exercises with invalid m/n and any remaining old-style exercises that haven't been migrated yet

Verification: all new and existing tests pass (test_contentnodes.py, test_views_internal.py).

References

Fixes #5667

Reviewer guidance

• utils/nodes.py:528-530 — the core fix: 3 lines that null out m/n for non-m_of_n models. This runs on every exercise API read/write via consolidate_extra_fields and ExtraFieldsSerializer.update, so correctness here is critical. Only m_of_n should preserve m/n values; all other mastery models should have them set to null.
• views/internal.py:839-842 — adding migrate_extra_fields to the ricecooker import path. This runs before the existing completion_criteria.validate() block, so newly-migrated data gets validated too. Test coverage confirms new-style extra_fields pass through unchanged.
• Management command — iterates all exercises using iterator() with Python-side filtering to avoid expensive nested JSON queries. Individual save(update_fields=...) per affected node. Safe for large datasets but will be slow on production (millions of exercises). Run with --dry-run first to verify counts.

[rtibbles]

This is the core of the fix, really - m and n should be None if we don't have an exercises.M_OF_N mastery model.

[rtibbles]

Updated!

[bjester]

LGTM! Tests have good coverage over critical logic. Dry run logic and command logging makes sense