feat(eval): GLM_WORKERS env knob — 8 default, lower for loaded endpoints

glm-4.7-flash currently sits behind sustained 429s (code 1305);
hammering it with 8 concurrent workers burns the 5-retry budget into
empty rows. The knob lets overloaded models run politely; empties
remain self-healing on resume (#65).

Author: ronaldtse

eval_sadeed_glm.py

@@ -132,7 +132,8 @@ def work(i: int) -> tuple[int, str]:
             return i, call(session, key, inputs[i])
 
         n_written = 0
-        with ThreadPoolExecutor(max_workers=8) as ex:
+        workers = int(os.environ.get("GLM_WORKERS", "8"))
+        with ThreadPoolExecutor(max_workers=workers) as ex:
             futures = {ex.submit(work, i): i for i in todo}
             for fut in as_completed(futures):
                 i, pred = fut.result()