fix: deps backfill fails due to missing log_record

This migration used to use the current models, which have a "log_record"
property in order to normalize things more between the Draft and
Published models, but this is not present in the historical models that
this migration was shifted to.

This fix uses the current Published/Draft verisons where that was
equivalent to checking the log records, and explicitly switches between
draft_log_record and publish_log_record in the places where we need to.

Increments version to 0.38.2

Author: ormsbee

src/openedx_content/backcompat/publishing/migrations/0010_backfill_dependencies.py

@@ -400,14 +400,23 @@ def hash_for_log_record(
     # Begin normal
     dependencies = sorted(
         record_ids_to_live_deps[record.id],
-        key=lambda entity: getattr(entity, branch).log_record.new_version_id,
+        key=lambda entity: getattr(entity, branch).version_id,
     )
     dep_state_entries = []
     for dep_entity in dependencies:
-        new_version_id = getattr(dep_entity, branch).log_record.new_version_id
+        new_version_id = getattr(dep_entity, branch).version_id
+
+        # Unfortunate inconsistency in field naming in the models. We have a
+        # property to normalize this, but that doesn't apply to the historical
+        # models.
+        if branch == "draft":
+            log_record = dep_entity.draft.draft_log_record
+        else:
+            log_record = dep_entity.published.publish_log_record
+
         hash_digest = hash_for_log_record(
             apps,
-            getattr(dep_entity, branch).log_record,
+            log_record,
             record_ids_to_hash_digests,
             record_ids_to_live_deps,
             untrusted_record_id_set,