[Test] Consolidate paged kernel tests into deterministic golden test (#167)

`test_metal_kernel_paged.py` re-implements vllm-metal internals (cache
setup, prefill/decode orchestration, context management) to compare two
paths. This scaffolding introduces additional complexity, making
failures hard to attribute.

Delete it and add its prompts to `test_paged_deterministic.py`, which
does the same comparison end-to-end through the real vLLM stack against
golden tokens.

Related: 
#158 
#149
#119

---------

Signed-off-by: ran <hzz5361@psu.edu>

Author: WindChimeRan

tests/test_metal_kernel_paged.py

@@ -46,6 +46,7 @@
     "One plus one equals",
     "The largest planet in our solar system is",
     "Water boils at a temperature of",
+    "Machine learning is",
 ]
 
 # fmt: off
@@ -57,6 +58,7 @@
     "One plus one equals":                        [825, 11, 825, 5519, 825, 16819, 1378, 13, 2055, 11],
     "The largest planet in our solar system is":  [1112, 30, 362, 13, 43562, 425, 13, 48976, 356, 13],
     "Water boils at a temperature of":            [220, 16, 15, 15, 30937, 13, 3555, 374, 279, 9315],
+    "Machine learning is":                         [264, 7988, 5392, 429, 702, 13791, 1506, 279, 2070, 315],
 }
 
 # Golden token IDs from paged KV cache (HF kernel on main branch), greedy decoding.
@@ -68,6 +70,7 @@
     "One plus one equals":                        [825, 11, 825, 5519, 825, 16819, 1378, 13, 2055, 11],
     "The largest planet in our solar system is":  [1112, 30, 362, 13, 43562, 425, 13, 48976, 356, 13],
     "Water boils at a temperature of":            [220, 16, 15, 15, 30937, 13, 3555, 374, 279, 9315],
+    "Machine learning is":                         [264, 7988, 5392, 429, 702, 13791, 1506, 279, 2070, 315],
 }
 # fmt: on
 

tests/test_paged_deterministic.py

@@ -32,6 +32,7 @@
     "One plus one equals",
     "The largest planet in our solar system is",
     "Water boils at a temperature of",
+    "Machine learning is",
 ]
 
 if __name__ == "__main__":