Docs: remove stale drf-writable-nested references, DRY up audit-actor update()

Final-review cleanup on refactor/remove-drf-writable-nested: several
comments/docstrings still described drf-writable-nested's removed
NestedUpdateMixin behavior in the present tense. Rewrote them to explain
the shared context-cache and delete-query-count guarantees in terms of
the current code (bulk-edit's per-field CustomFieldInstanceSerializer
construction, and _sync_custom_fields' single hard-delete query).

Also collapsed the duplicated `if custom_fields_data is not None:
self._sync_custom_fields(...)` line in DocumentSerializer.update()'s
audit-log branches into a single code path using
contextlib.nullcontext(), removing the drift risk that caused the
earlier audit-actor bug.

No behavior change.

src/documents/serialisers.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import contextlib
 import logging
 import math
 import re
@@ -956,19 +957,19 @@ def validate_documentlink_targets(user, doc_ids):
         )
 
 
-# drf-writable-nested revalidates a document's custom_fields more than once
+# A CustomField lookup cache scoped to a single field/serializer instance
-# per request: once as the ordinary nested list, then again per-item while
+# only helps within that one instance's own validation pass. Several call
-# matching existing vs. new CustomFieldInstance rows during save() -- and
+# sites, though, build more than one CustomFieldInstanceSerializer (or its
-# that second pass builds a brand new serializer (and field) instance per
+# CustomFieldInstanceListSerializer/field) for the same request and pass
-# item (see its update_or_create_reverse_relations / _get_serializer_for_field),
+# each of them `context=self.context` -- the *same* dict object, not a
-# so a cache on the field instance alone only helps the first pass. It does,
+# copy -- e.g. bulk-edit's _validate_custom_field_values() constructing a
-# however, explicitly pass `context=self.context` to every one of those
+# fresh CustomFieldInstanceSerializer per submitted field. That context
-# fresh serializers -- the *same* dict object the outer DocumentSerializer
+# dict is already request-scoped (DRF builds it fresh per request via
-# is using, not a copy. That context dict is already request-scoped (DRF
+# get_serializer_context()), so stashing the resolved CustomField objects
-# builds it fresh per request via get_serializer_context()), so stashing the
+# there -- rather than in some new global/thread-local cache -- lets every
-# resolved CustomField objects there -- rather than in some new global/
+# one of those separately-instantiated serializers reuse them for free
-# thread-local cache -- lets every later pass reuse them for free while
+# while staying entirely within DRF's existing, already-request-scoped
-# staying entirely within DRF's existing, already-request-scoped machinery.
+# machinery.
 _CUSTOM_FIELD_CONTEXT_CACHE_KEY = "_custom_field_lookup_cache"
 
 
@@ -1438,12 +1439,12 @@ class DocumentSerializer(
 
         custom_fields_data = validated_data.pop("custom_fields", None)
 
-        if settings.AUDIT_LOG_ENABLED:
+        actor_context = (
-            with set_actor(self.user):
+            set_actor(self.user)
-                super().update(instance, validated_data)
+            if settings.AUDIT_LOG_ENABLED
-                if custom_fields_data is not None:
+            else contextlib.nullcontext()
-                    self._sync_custom_fields(instance, custom_fields_data)
+        )
-        else:
+        with actor_context:
             super().update(instance, validated_data)
             if custom_fields_data is not None:
                 self._sync_custom_fields(instance, custom_fields_data)

src/documents/tests/test_api_custom_fields.py

@@ -592,9 +592,10 @@ class TestCustomFieldsAPI(DirectoriesMixin, APITestCase):
         WHEN:
             - A second, separately-instantiated CustomFieldInstanceSerializer
               validates the same field id, sharing that same context
-              (this is what drf-writable-nested does: it rebuilds a fresh
+              (this mirrors what real call sites do, e.g. bulk-edit's
-              serializer -- and fresh field instances -- per item while
+              validate_custom_fields()/_validate_custom_field_values()
-              matching existing vs. new instances during save())
+              constructing a fresh CustomFieldInstanceSerializer per
+              submitted field, all sharing the outer serializer's context)
         THEN:
             - No additional query is issued to resolve the CustomField
         """
@@ -1560,12 +1561,10 @@ class TestCustomFieldsAPI(DirectoriesMixin, APITestCase):
             - A PATCH request updates 2 of them, adds 1 new one, and omits
               the 3rd (which should be deleted)
         THEN:
-            - The number of queries used to create/update/delete the
+            - The omitted custom field is removed via exactly one
-              CustomFieldInstance rows is bounded and does not scale with
+              delete-shaped query, not drf-writable-nested's old two-step
-              drf-writable-nested's per-item serializer-rebuild + separate
+              soft-delete-then-hard-delete-after pattern (two delete
-              soft-delete-then-hard-delete-after pattern (one query per
+              passes)
-              CustomFieldInstance write via update_or_create, plus exactly
-              one final hard-delete query -- not two delete passes)
         """
         doc = Document.objects.create(
             title="WOW",