chore: set DEFER_PYDANTIC_BUILD in deployment instead of settings.py

The DEFER_PYDANTIC_BUILD=0 workaround for the OpenAI SDK MockValSer error is
applied manually in the deployment config, so drop the os.environ.setdefault
from config/settings.py. Keep the documented entry in .env.example (without the
now-removed settings.py reference). See openai/openai-python#1306.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: SmittieC

.env.example

@@ -129,9 +129,9 @@ APP_IP=
 ## providers and by integration tests in apps/service_providers/tests/test_llm_integration.py).
 ## Set the keys for whichever providers you want available; unset providers are skipped.
 
-## Forces the OpenAI SDK to build its Pydantic models eagerly instead of deferring, which otherwise
-## causes "TypeError: 'MockValSer' object is not an instance of 'SchemaSerializer'" when serializing
-## responses. Defaulted to "0" in config/settings.py; only set here to override. See openai/openai-python#1306.
+## Set to "0" to force the OpenAI SDK to build its Pydantic models eagerly instead of deferring,
+## which otherwise causes "TypeError: 'MockValSer' object is not an instance of 'SchemaSerializer'"
+## when serializing responses. See openai/openai-python#1306.
 # DEFER_PYDANTIC_BUILD=0
 
 # OpenAI

config/settings.py

@@ -27,13 +27,6 @@
 env = environ.Env()
 env.read_env(os.path.join(BASE_DIR, ".env"))
 
-# The OpenAI SDK defers building its Pydantic models by default (defer_build=True), which leaves
-# nested/generic response models with a MockValSer placeholder serializer. Serializing such a
-# response (e.g. langchain_openai calling response.model_dump() on a ParsedResponse) then raises
-# "TypeError: 'MockValSer' object is not an instance of 'SchemaSerializer'". Building eagerly avoids
-# this. Must be set before `openai` is first imported, hence here. See openai/openai-python#1306.
-os.environ.setdefault("DEFER_PYDANTIC_BUILD", "0")
-
 # Quick-start development settings - unsuitable for production
 # See https://docs.djangoproject.com/en/3.2/howto/deployment/checklist/