feat: add P7 OCR soak scenario and scoreboard

tools/soak/scenarios.py

@@ -1,8 +1,9 @@
-"""P1-P6 场景请求生成器(findings §2 矩阵;async 生成器,产出调用参数)。
+"""P1-P7 场景请求生成器(findings §2 矩阵;async 生成器,产出调用参数)。
 
-每项产出 `(kind, kwargs)`: kind ∈ {"chat"};kwargs 直接喂
-`GatewayClient.chat(**kwargs)`。回放/组装场景一律掺 `cache_salt=run_id`
-破缓存(缓存行为归 P4);P4 子流量特意重复 messages 且不掺 salt。
+每项产出 `(kind, kwargs)`: kind ∈ {"chat", "ocr_text", "ocr_layout"};
+chat 喂 `GatewayClient.chat(**kwargs)`,ocr_* 喂 `OcrClient` 对应方法
+(M3 计划 T8)。回放/组装场景一律掺 `cache_salt=run_id` 破缓存(缓存
+行为归 P4);P4 子流量特意重复 messages 且不掺 salt。
 """
 
 from __future__ import annotations
@@ -10,10 +11,13 @@ from __future__ import annotations
 import random
 from dataclasses import dataclass
 from pathlib import Path
-from typing import Any, AsyncIterator
+from typing import TYPE_CHECKING, Any
 
 from pydantic import BaseModel
 
+if TYPE_CHECKING:
+    from collections.abc import AsyncIterator
+
 from tools.soak.corpus import (
     assemble_frame_messages,
     chs_image_messages,
@@ -182,6 +186,19 @@ async def p6_mixed_soak(
             }[key](corpus, f"{run_id}-r{rng()}", rng)
 
 
+async def p7_ocr(
+    corpus: SoakCorpus, run_id: str, rng=random.random, *, layout_ratio: float = 0.2
+) -> AsyncIterator[Item]:
+    """P7 OCR 双端点混合(M3 计划 T8): 真实图像语料循环,text/layout 8:2。
+
+    故障性来自 SOAK_OCR 源池配置(黑洞/坏端口),生成器本身不造故障。
+    """
+    while True:
+        image = corpus.images[int(rng() * len(corpus.images)) % len(corpus.images)]
+        kind = "ocr_layout" if rng() < layout_ratio else "ocr_text"
+        yield (kind, {"image": image.read_bytes(), "session_id": f"{run_id}-p7"})
+
+
 SCENARIOS = {
     "P1": p1_trace_chains,
     "P2": p2_multimodal_replay,
@@ -189,4 +206,5 @@ SCENARIOS = {
     "P4": p4_cache_bidirectional,
     "P5": p5_fault_mixed,
     "P6": p6_mixed_soak,
+    "P7": p7_ocr,
 }