fix: rename benchmark for CI discovery, bump to 10k iterations

- Renamed benchmark_ngram_kernel.py → test_benchmark_ngram_kernel.py
  so pytest discovers it (test_*.py pattern)
- Bumped NUM_ITERS 10→10000, WARMUP 2→5 for noise-free profiling
- Gated benchmark class with RUN_NGRAM_BENCHMARKS=1 (won't bloat CI)

Author: cloudforge1

…ts/spec_decode/benchmark_ngram_kernel.py …ec_decode/test_benchmark_ngram_kernel.pytests/spec_decode/benchmark_ngram_kernel.py renamed to tests/spec_decode/test_benchmark_ngram_kernel.py tests/spec_decode/benchmark_ngram_kernel.py renamed to tests/spec_decode/test_benchmark_ngram_kernel.py

@@ -40,8 +40,8 @@
 
 MAX_NGRAM_SIZE = 3
 MAX_DRAFT_TOKENS = 10
-NUM_ITERS = 10
-WARMUP = 2
+NUM_ITERS = 10000
+WARMUP = 5
 
 
 def _build_data(batch_size, seq_len, hit_type="low_input", seed=42):
@@ -206,6 +206,10 @@ def _print_table(title, header, rows):
     print(f"{'=' * 80}")
 
 
+@unittest.skipUnless(
+    os.environ.get("RUN_NGRAM_BENCHMARKS", "0") == "1",
+    "Set RUN_NGRAM_BENCHMARKS=1 to run multi-group profiling (slow)",
+)
 class TestNgramBenchmarkGroups(unittest.TestCase):
     """Multi-dimension benchmark matching NKNaN's 5-group methodology."""