fix(crawl): throttle concurrent CrawlJobs and relax fpw/proxyscrape HTTP

- CrawlJob waits on crawl_slot before JobExecutor semaphore so crawl-all does not fill slots while queued
- BaseHTTPPlugin: longer connect budget for slow international links
- proxyscrape: jsDelivr mirror + longer GitHub/API phases
- fpw_*: higher timeouts/retries; lower internal concurrency on heavy multi-URL plugins

Made-with: Cursor

app/core/execution/crawl_gate.py

@@ -0,0 +1,22 @@
+"""批量爬取时限制同时发起 HTTP 的插件数，避免 crawl-all 与验证/聚合任务抢满执行器槽位。"""
+import asyncio
+from contextlib import asynccontextmanager
+from typing import AsyncIterator
+
+# 与单插件内 max_concurrency 相乘后仍应对外网友好；过小会拉长总耗时。
+CRAWL_MAX_CONCURRENT = 4
+
+_sem: asyncio.Semaphore | None = None
+
+
+def _get_sem() -> asyncio.Semaphore:
+    global _sem
+    if _sem is None:
+        _sem = asyncio.Semaphore(CRAWL_MAX_CONCURRENT)
+    return _sem
+
+
+@asynccontextmanager
+async def crawl_slot() -> AsyncIterator[None]:
+    async with _get_sem():
+        yield