fix: align analytics message count with token count by filtering on assistant role

The get_message_count_by_user query was counting messages of ALL roles
(user, assistant, system), while get_token_usage_by_user only aggregated
tokens from assistant messages with non-null usage data.

This mismatch caused the Admin Panel Analytics to show inflated message
counts alongside zero tokens for users whose messages were predominantly
user-role messages without any token usage data.

Align the message count query with the token usage query by adding a
role='assistant' filter, so both metrics reflect the same underlying
dataset of assistant responses.

backend/open_webui/models/chat_messages.py

@@ -478,7 +478,10 @@ class ChatMessageTable:
 
             stmt = (
                 select(ChatMessage.user_id, func.count(ChatMessage.id).label('count'))
-                .filter(~ChatMessage.user_id.like('shared-%'))
+                .filter(
+                    ChatMessage.role == 'assistant',
+                    ~ChatMessage.user_id.like('shared-%'),
+                )
             )
 
             if start_date: