fix(proxy): tolerate a SpendLogsPayload without total_tokens

Reading payload["total_tokens"] unconditionally aborted _batch_database_updates
with a KeyError when a (test or partial) payload omits the key, dropping the
rest of the batch. Guard the typed index with a membership check so a missing
count falls back to 0, staying tolerant like the old .get while keeping the
read type-clean.

litellm/proxy/db/db_spend_update_writer.py

@@ -1066,7 +1066,7 @@ class DBSpendUpdateWriter:
                 team_id=team_id,
                 user_id=user_id,
                 prisma_client=prisma_client,
-                tokens=payload["total_tokens"],
+                tokens=payload["total_tokens"] if "total_tokens" in payload else 0,
             )
         except Exception:
             verbose_proxy_logger.debug(