[CI/Build][AMD] Use float16 in test_reset_prefix_cache_e2e to avoid accuracy issues (#29997)

Signed-off-by: Randall Smith <ransmith@amd.com>
Co-authored-by: Randall Smith <ransmith@amd.com>

tests/v1/core/test_reset_prefix_cache_e2e.py

@@ -21,6 +21,7 @@ def test_reset_prefix_cache_e2e(monkeypatch):
         max_num_batched_tokens=32,
         max_model_len=2048,
         compilation_config={"mode": 0},
+        dtype="float16",
     )
     engine = LLMEngine.from_engine_args(engine_args)
     sampling_params = SamplingParams(