feat(write_attachments): add --append mode, fix batch timing and SQL Server param limit

- Add --append flag: diff source-table 总排号 against existing
  Common.Attachment SN and classify/write only the missing ones. --limit
  caps the per-run append count, --order sets the direction. Backed by
  new fetch_existing_attachment_sns() in db.py.
- Fix batch elapsed-time accounting: summarize_results() summed each
  task's per-item elapsed_ms, which overcounts under ThreadPoolExecutor
  concurrency (cumulative work time, not real wall-clock — 100 tasks on
  8 workers reported ~5x the actual runtime). Callers now time
  classify_batch() via perf_counter and pass wall_clock_ms; both the
  write_attachments [汇总] line and main.py --summary report wall-clock
  separately from the cumulative sum.
- Format durations >=1s in seconds (88851.4 ms -> 88.85 s) in the
  human-readable [汇总] line; structured JSON --summary fields stay in ms.
- Chunk all IN (...) lists to 2000 items to respect SQL Server's 2100
  bind-parameter hard limit (previously --append --limit 5000 failed at
  the fetch step with "COUNT 字段不正确"). Applied to fetch_params_by_ids
  (sn / id paths) and to the DELETE inside upsert_attachments.

Co-Authored-By: Claude <noreply@anthropic.com>

classifier.py

@@ -265,12 +265,21 @@ def classify_single(
     )[0]
 
 
-def summarize_results(results: list[dict[str, Any]]) -> dict[str, Any]:
+def summarize_results(
+    results: list[dict[str, Any]],
+    wall_clock_ms: float | None = None,
+) -> dict[str, Any]:
     """汇总一批 classify_batch 结果的运行统计。
 
-    从每条结果的 meta 里累加：总耗时、实际 LLM 调用次数、各类 token 总量、
-    以及整体缓存命中率（按 prompt_cache_hit_tokens / prompt_tokens 加权）。
-    供 CLI 打印批统计用，不改变 classify_batch 的返回结构。
+    从每条结果的 meta 里累加：各结果 elapsed_ms 的总和、实际 LLM 调用次数、
+    各类 token 总量，以及整体缓存命中率（按 prompt_cache_hit_tokens /
+    prompt_tokens 加权）。供 CLI 打印批统计用，不改变 classify_batch 的返回结构。
+
+    注意耗时两个字段的区别（classify_batch 是并发执行的）：
+      - total_elapsed_ms：各结果自身耗时(elapsed_ms)的**累加**，即"累计工作量"
+        （相当于把这些任务串行跑的总时长），并发下会明显大于真实墙钟；
+      - wall_clock_ms：调用方传入的整批分类**真实墙钟**耗时（在 classify_batch
+        外层用 perf_counter 计时），二者之比 ≈ 并发增益（≈ max_workers）。
     """
     total_elapsed = 0.0
     llm_calls = 0
@@ -291,7 +300,7 @@ def summarize_results(results: list[dict[str, Any]]) -> dict[str, Any]:
                 sum_hit += u.get("prompt_cache_hit_tokens", 0) or 0
                 sum_miss += u.get("prompt_cache_miss_tokens", 0) or 0
     avg_cache_hit_rate = round(sum_hit / sum_prompt, 4) if sum_prompt else 0.0
-    return {
+    out = {
         "count": len(results),
         "total_elapsed_ms": round(total_elapsed, 1),
         "llm_calls": llm_calls,
@@ -302,3 +311,6 @@ def summarize_results(results: list[dict[str, Any]]) -> dict[str, Any]:
         "total_cache_miss_tokens": sum_miss,
         "avg_cache_hit_rate": avg_cache_hit_rate,
     }
+    if wall_clock_ms is not None:
+        out["wall_clock_ms"] = round(float(wall_clock_ms), 1)
+    return out