fix: count actual tool calls instead of tool-related messages

tool_call_count was inaccurate in two ways:

1. Under-counting: an assistant message with N parallel tool calls
   (e.g. "kill the light and shut off the fan" = 2 ha_call_service)
   only incremented tool_call_count by 1 instead of N.

2. Over-counting: tool response messages (role=tool) also incremented
   tool_call_count, double-counting every tool interaction.

Combined: 2 parallel tool calls produced tool_call_count=3 (1 from
assistant + 2 from tool responses) instead of the correct value of 2.

Fix: only count from assistant messages with tool_calls, incrementing
by len(tool_calls) to handle parallel calls correctly. Tool response
messages no longer affect tool_call_count.

This impacts /insights and /usage accuracy for sessions with tool use.

tests/test_hermes_state.py

@@ -94,13 +94,50 @@ class TestMessageStorage:
         session = db.get_session("s1")
         assert session["message_count"] == 2
 
-    def test_tool_message_increments_tool_count(self, db):
+    def test_tool_response_does_not_increment_tool_count(self, db):
+        """Tool responses (role=tool) should not increment tool_call_count.
+
+        Only assistant messages with tool_calls should count.
+        """
         db.create_session(session_id="s1", source="cli")
         db.append_message("s1", role="tool", content="result", tool_name="web_search")
 
+        session = db.get_session("s1")
+        assert session["tool_call_count"] == 0
+
+    def test_assistant_tool_calls_increment_by_count(self, db):
+        """An assistant message with N tool_calls should increment by N."""
+        db.create_session(session_id="s1", source="cli")
+        tool_calls = [
+            {"id": "call_1", "function": {"name": "web_search", "arguments": "{}"}},
+        ]
+        db.append_message("s1", role="assistant", content="", tool_calls=tool_calls)
+
         session = db.get_session("s1")
         assert session["tool_call_count"] == 1
 
+    def test_tool_call_count_matches_actual_calls(self, db):
+        """tool_call_count should equal the number of tool calls made, not messages."""
+        db.create_session(session_id="s1", source="cli")
+
+        # Assistant makes 2 parallel tool calls in one message
+        tool_calls = [
+            {"id": "call_1", "function": {"name": "ha_call_service", "arguments": "{}"}},
+            {"id": "call_2", "function": {"name": "ha_call_service", "arguments": "{}"}},
+        ]
+        db.append_message("s1", role="assistant", content="", tool_calls=tool_calls)
+
+        # Two tool responses come back
+        db.append_message("s1", role="tool", content="ok", tool_name="ha_call_service")
+        db.append_message("s1", role="tool", content="ok", tool_name="ha_call_service")
+
+        session = db.get_session("s1")
+        # Should be 2 (the actual number of tool calls), not 3
+        assert session["tool_call_count"] == 2, (
+            f"Expected 2 tool calls but got {session['tool_call_count']}. "
+            "tool responses are double-counted and multi-call messages are under-counted"
+        )
+
     def test_tool_calls_serialization(self, db):
         db.create_session(session_id="s1", source="cli")
         tool_calls = [{"id": "call_1", "function": {"name": "web_search", "arguments": "{}"}}]