[Feature][V1]: suupports cached_tokens in response usage (#18149)

Co-authored-by: simon-mo <xmo@berkeley.edu>

tests/v1/core/test_scheduler_e2e.py

@@ -19,7 +19,8 @@ def model() -> LLM:
                enable_prefix_caching=True,
                long_prefill_token_threshold=2,
                max_num_batched_tokens=6,
-               max_num_seqs=3)
+               max_num_seqs=3,
+               block_size=16)
 
 
 def test_concurrent_partial_prefill(model):
@@ -27,3 +28,11 @@ def test_concurrent_partial_prefill(model):
     assert len(outputs) == 3
     for output in outputs:
         assert len(output.outputs) == 1
+
+
+def test_prefix_cache_stats_is_recorded(model):
+    # 17 tokens will make sure first 16 tokens are cached in a block
+    input_tokens = {"prompt_token_ids": [101] * 17}
+    _ = model.generate([input_tokens])
+    outputs = model.generate([input_tokens])
+    assert outputs[0].num_cached_tokens == 16