fix(settings): make DATA_UPLOAD_MAX_MEMORY_SIZE env-configurable

Previously hardcoded at 100 MB in base.py. In practice ADC workers
post ML result payloads for a full batch (detection coordinates +
classifications for tens of images) in a single POST, and those have
been observed in the 139–321 MB range on staging for the
global_moths_2024 pipeline — well above the 100 MB ceiling.

Raising the limit in code would be both premature (proper fix is
worker-side incremental posting, tracked in #1223) and environment-
specific (staging may need to tolerate today's payloads; production
may want a tighter ceiling to catch regressions). Making it an env
override lets each deployment tune without a code change and without
maintaining a hot-patch on the server.

Reads from ``DJANGO_DATA_UPLOAD_MAX_MEMORY_MB`` (integer, in MB).
Default stays at 100 MB so existing deployments see no change unless
they opt in.

Nginx's ``client_max_body_size`` still needs to be raised in lockstep
on the fronting proxy — that is independently configurable and lives
outside this repo.

Author: mihow

config/settings/base.py

@@ -426,8 +426,15 @@ def _celery_result_backend_url(redis_url):
 CELERY_BROKER_CONNECTION_MAX_RETRIES = None  # Retry forever
 
 
-# Allow large request bodies from ML workers posting classification results
-DATA_UPLOAD_MAX_MEMORY_SIZE = 100 * 1024 * 1024  # 100MB (default 2.5MB)
+# Allow large request bodies from ML workers posting classification results.
+# ML detection+classification payloads for a single batch can easily exceed
+# the Django default (2.5 MB) and even the previous hardcoded 100 MB ceiling.
+# Configurable via env (MB) so staging and production can tune without a
+# code change. See RolnickLab/antenna#1223 for the longer-term fix (worker-
+# side incremental result posting).
+DATA_UPLOAD_MAX_MEMORY_SIZE = (
+    env.int("DJANGO_DATA_UPLOAD_MAX_MEMORY_MB", default=100) * 1024 * 1024  # type: ignore[no-untyped-call]
+)
 
 # django-rest-framework
 # -------------------------------------------------------------------------------