feat: HomeFormatter parses inline entity tags instead of tool_end blocks

The supervisor LLM now embeds <type>[id1,id2]</type> entity tags in its
response text. The HomeFormatter buffers streamed tokens, detects complete
tags across chunk boundaries, and emits WsStreamBlock with entity type +
specific IDs. This replaces the old approach of emitting blocks for every
tool_end event, which dumped ALL entities regardless of relevance.

Also fixes:
- NoneType guard on metadata in _run_graph_stream (metadata can be None)
- Updated _HOME_SYSTEM prompt with entity tag instructions
- Updated all affected tests

app/core/deep_agent.py

@@ -236,7 +236,19 @@ _HOME_SYSTEM = (
     "multiple sub-agents if needed.\n\n"
     "You also have an update_core_memory tool — use it when the user states "
     "a preference or important fact worth remembering long-term.\n\n"
-    "After gathering data, synthesize a clear, helpful response for the user.\n\n"
+    "## Entity References\n"
+    "When your response mentions specific workspace entities, embed them "
+    "inline using entity tags so the UI can render interactive components.\n"
+    "Format: <type>[comma-separated UUIDs]</type>\n"
+    "Supported types: task, project, note, timeline\n\n"
+    "Example response:\n"
+    "  Here is your project:\n"
+    "  <project>[abc-123-def]</project>\n"
+    "  It has these pending tasks:\n"
+    "  <task>[def-456,ghi-789]</task>\n\n"
+    "IMPORTANT: Only include IDs of entities that are directly relevant to "
+    "the user's question. Do NOT dump all entity IDs returned by a tool — "
+    "filter to only the ones the user asked about or that matter for the answer.\n\n"
     "Memory context:\n{memory_context}"
 )
 
@@ -360,6 +372,7 @@ async def _run_graph_stream(
                     isinstance(msg, AIMessageChunk)
                     and msg.content
                     and not msg.tool_calls
+                    and isinstance(metadata, dict)
                     and metadata.get("langgraph_node") == "agent"
                 ):
                     yield ("token", str(msg.content))