hive-runtime: report an empty ACP end_turn as a turn error (#4819)

opencode answers `end_turn` even when its provider rejected the request
with a non-retryable error (401, 400), and forwards nothing over ACP, so
the turn looked like an empty success. A turn that ends with `end_turn`,
no event, no `usage_update` and no `usage` in the prompt response now
fails with `AcpError::EmptyEndTurn`.

A turn that really produced nothing and reported no usage is reported
the same way; that false positive is accepted.

The test agent's plain `end_turn` replies now carry a response `usage`,
so its ordinary turns stay successes; a response `usage` feeds only cost
telemetry, not the compaction watermark. A new `blank` mode keeps the
empty reply for the error case.

hive-runtime/src/acp/stream.rs

@@ -21,6 +21,7 @@ pub(super) struct StreamMapper {
     thought: String,
     tools: HashMap<String, ToolCall>,
     usage: Option<(u64, u64)>,
+    emitted: bool,
 }
 
 struct ToolCall {
@@ -39,6 +40,7 @@ impl StreamMapper {
             thought: String::new(),
             tools: HashMap::new(),
             usage: None,
+            emitted: false,
         }
     }
 
@@ -69,6 +71,7 @@ impl StreamMapper {
             }
             _ => {}
         }
+        self.emitted |= !out.is_empty();
         out
     }
 
@@ -82,9 +85,20 @@ impl StreamMapper {
         for (id, tool) in pending {
             out.push(tool_use(&id, &tool.name, &tool.input));
         }
+        self.emitted |= !out.is_empty();
         out
     }
 
+    /// Whether the turn has emitted any event.
+    pub(super) fn has_content(&self) -> bool {
+        self.emitted
+    }
+
+    /// Whether the turn has had a `usage_update`.
+    pub(super) fn has_usage_update(&self) -> bool {
+        self.usage.is_some()
+    }
+
     /// The turn's telemetry: context from the last `usage_update`, cost from
     /// the `session/prompt` response's `usage` when the agent sends one.
     pub(super) fn telemetry(&self, response: &Value, model: Option<&str>) -> Telemetry {