studio: drop redundant boundary flag on post-tool status, do reset at tool_end

Codex 02:42Z on #5549 caught that flagging the post-tool empty-status
event with boundary=True (cycle-12 commit 610c387) double-handles the
cursor reset in the Anthropic streaming path.

AnthropicStreamEmitter._handle_tool_end already:
  - closes the open tool_use block,
  - emits tool_result,
  - increments block_index,
  - opens a fresh text block, and
  - resets _prev_text = "".

When llama_cpp.py then yielded boundary=True on the very next event,
_handle_boundary fired _close_block + _open_text_block on that freshly
opened (still-empty) text block. Result: every tool call produced a
spurious content_block_stop + content_block_start pair before the
post-tool model text streamed.

Fix:

  - llama_cpp.py: drop boundary=True from the post-tool status emit
    (line ~4768). Keep boundary=True only at the auto-continue site
    (line ~4500), which has no preceding tool_end to do the cursor
    work.
  - routes/inference.py OpenAI-compat tool stream: mirror the
    Anthropic semantics by resetting prev_text on BOTH tool_start AND
    tool_end, so the post-tool empty-status no longer needs to do it.

Add backend/tests/test_anthropic_messages.py::TestAnthropicStreamEmitter::
test_post_tool_empty_status_does_not_double_close as a regression
test: content -> tool_start -> tool_end -> empty status -> content
must not bump block_index past tool_end's increment, and the post-tool
content must land in the text block tool_end opened.

95 tests pass across the anthropic + trailing-plan suites.

studio/backend/core/inference/llama_cpp.py

@@ -4757,9 +4757,15 @@ class LlamaCppBackend:
                     conversation.append(tool_msg)
 
                 # Clear tool status badge before next generation iteration.
-                # boundary=True: the model is about to start a fresh turn
-                # so cumulative-text adapters must reset their cursor.
-                yield {"type": "status", "text": "", "boundary": True}
+                # We do NOT mark this as a boundary: the preceding
+                # tool_end event already opened a fresh text block and
+                # reset the cumulative cursor in Anthropic streaming
+                # (see AnthropicStreamEmitter._handle_tool_end), and the
+                # OpenAI-compat path resets prev_text on tool_start. A
+                # second boundary here would close the freshly opened
+                # text block (empty) and reopen it, producing a spurious
+                # content_block_stop/start pair on every tool call.
+                yield {"type": "status", "text": ""}
                 # Continue the loop to let model respond with context
                 continue
 

studio/backend/routes/inference.py

@@ -2481,8 +2481,18 @@ async def openai_chat_completions(
                             continue
 
                         if event["type"] in ("tool_start", "tool_end"):
-                            if event["type"] == "tool_start":
-                                prev_text = ""
+                            # Both endpoints of a tool call begin a fresh
+                            # cumulative-text window: tool_start because
+                            # the model's next visible content restarts
+                            # cumulative-from-zero in the post-tool turn,
+                            # and tool_end because after we emit the
+                            # tool result, the model's next iteration
+                            # produces its own fresh cumulative stream
+                            # (the post-tool empty-status event is just
+                            # a UI badge clear; it does NOT carry a
+                            # boundary flag, so this is the place to
+                            # reset prev_text on tool_end).
+                            prev_text = ""
                             yield f"data: {json.dumps(event)}\n\n"
                             continue
 

studio/backend/tests/test_anthropic_messages.py

@@ -649,6 +649,47 @@ class TestAnthropicStreamEmitter:
         parsed = json.loads(nxt[0].split("data: ")[1])
         assert parsed["delta"]["text"] == "second turn"
 
+    def test_post_tool_empty_status_does_not_double_close(self):
+        """After tool_end already opens a fresh text block, the post-tool
+        empty-status event emitted by llama_cpp.py (line 4766) must NOT
+        close that fresh block. Otherwise every tool call produces a
+        spurious empty content_block_stop + content_block_start pair
+        before the model's post-tool text arrives. Regression test for
+        PR 5549 codex 02:42Z."""
+
+        e = AnthropicStreamEmitter()
+        e.start("msg_1", "m")
+        e.feed({"type": "content", "text": "pre"})
+        e.feed(
+            {
+                "type": "tool_start",
+                "tool_name": "t",
+                "tool_call_id": "tc_1",
+                "arguments": {},
+            }
+        )
+        e.feed(
+            {
+                "type": "tool_end",
+                "tool_name": "t",
+                "tool_call_id": "tc_1",
+                "result": "ok",
+            }
+        )
+        block_after_tool = e.block_index
+        # Post-tool empty status (no boundary flag): should produce zero
+        # SSE events and leave block_index unchanged. The previous
+        # behaviour was to close+reopen, which produced a duplicate
+        # empty content block.
+        out = e.feed({"type": "status", "text": ""})
+        assert out == []
+        assert e.block_index == block_after_tool
+        # Next content delta lands in the same fresh text block that
+        # tool_end opened.
+        nxt = e.feed({"type": "content", "text": "post"})
+        parsed = json.loads(nxt[0].split("data: ")[1])
+        assert parsed["delta"]["text"] == "post"
+
     def test_empty_status_without_boundary_does_not_close_block(self):
         """A non-boundary empty-status event (UI badge clear at normal
         stream end, draining fallbacks, final status yields in