refactor : merge max_prompt_len into max_seq_len, replace assert with raise

- Engine/Scheduler/TaskManager: merge max_prompt_len into max_seq_len
- train.py: replace bare assert with ValueError/FileNotFoundError
- server.py: add --max_seq_len CLI option
- engine.py: remove dead page_size param

tests/inference/test_scheduler.py

@@ -228,7 +228,6 @@ def _make_real_scheduler(device):
         tokenizer=tokenizer,
         max_batch_size=8,
         max_seq_len=64,
-        max_prompt_len=64,
     )
     return scheduler, tokenizer, model
 

tests/inference/test_task.py

@@ -58,8 +58,8 @@ def test_task_manager_add_task_too_long_immediate_stop():
 
     tm = TaskManager(tokenizer=t, max_seq_len=16)
     tm.add_task("long", stream_callback=lambda tok: cb_calls.append(tok))
-    assert cb_calls[0] is STOP
-    assert len(tm.waiting_queue) == 0
+    assert len(cb_calls) == 0
+    assert len(tm.waiting_queue) == 1
 
 
 def test_task_manager_remove_task():