fix: use max_completion_tokens instead of max_tokens for GPT-4o

Azure OpenAI GPT-4o and newer models require 'max_completion_tokens'
instead of 'max_tokens'. Fixed in all 3 services:
- chat_service.py (2 call sites)
- azure_ai.py (validation service)
- tenants/router.py (test connection endpoint)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

backend/app/services/azure_ai.py

@@ -124,7 +124,7 @@ def _call_azure_vision(
                     ],
                 }
             ],
-            max_tokens=max_tokens,
+            max_completion_tokens=max_tokens,
             temperature=temperature,
         )