HACK: workaround triton kernel compile failure

With:

```
$ VLLM_LOGGING_LEVEL=DEBUG VLLM_USE_V1=1 VLLM_ENABLE_V1_MULTIPROCESSING=1 vllm serve meta-llama/Llama-3.1-8B-Instruct --speculative-model "[ngram]" --num-speculative-tokens 5 --ngram-prompt-lookup-max 4
```

I'm getting:

```
  File "/home/markmc/vllm-project/vllm/vllm/v1/engine/core.py", line 121, in _initialize_kv_caches
    available_gpu_memory = self.model_executor.determine_available_memory()
                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/markmc/vllm-project/vllm/vllm/v1/executor/abstract.py", line 66, in determine_available_memory
    output = self.collective_rpc("determine_available_memory")
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/markmc/vllm-project/vllm/vllm/executor/uniproc_executor.py", line 56, in collective_rpc
    answer = run_method(self.driver_worker, method, args, kwargs)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/markmc/vllm-project/vllm/vllm/utils.py", line 2216, in run_method
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/markmc/vllm-project/vllm-venv/lib64/python3.12/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/markmc/vllm-project/vllm/vllm/v1/worker/gpu_worker.py", line 157, in determine_available_memory
    self.model_runner.profile_run()
  File "/home/markmc/vllm-project/vllm/vllm/v1/worker/gpu_model_runner.py", line 1466, in profile_run
    sampler_output = self._dummy_sampler_run(hidden_states)
                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/markmc/vllm-project/vllm-venv/lib64/python3.12/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/markmc/vllm-project/vllm/vllm/v1/worker/gpu_model_runner.py", line 1375, in _dummy_sampler_run
    self.rejection_sampler(
  File "/home/markmc/vllm-project/vllm-venv/lib64/python3.12/site-packages/torch/nn/modules/module.py", line 1736, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/markmc/vllm-project/vllm-venv/lib64/python3.12/site-packages/torch/nn/modules/module.py", line 1747, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/markmc/vllm-project/vllm/vllm/v1/sample/rejection_sampler.py", line 92, in forward
    output_token_ids = rejection_sample(
                       ^^^^^^^^^^^^^^^^^
  File "/home/markmc/vllm-project/vllm/vllm/v1/sample/rejection_sampler.py", line 166, in rejection_sample
    rejection_greedy_sample_kernel[(batch_size, )](
  File "/home/markmc/vllm-project/vllm-venv/lib64/python3.12/site-packages/triton/runtime/jit.py", line 345, in <lambda>
    return lambda *args, **kwargs: self.run(grid=grid, warmup=False, *args, **kwargs)
                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/markmc/vllm-project/vllm-venv/lib64/python3.12/site-packages/triton/runtime/jit.py", line 662, in run
    kernel = self.compile(
             ^^^^^^^^^^^^^
  File "/home/markmc/vllm-project/vllm-venv/lib64/python3.12/site-packages/triton/compiler/compiler.py", line 276, in compile
    module = src.make_ir(options, codegen_fns, context)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/markmc/vllm-project/vllm-venv/lib64/python3.12/site-packages/triton/compiler/compiler.py", line 113, in make_ir
    return ast_to_ttir(self.fn, self, context=context, options=options, codegen_fns=codegen_fns)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
triton.compiler.errors.CompilationError: at 17:7:
    bonus_token_ids_ptr,  # [batch_size]
    is_greedy_ptr,  # [batch_size] or None
    max_spec_len,
):
    req_idx = tl.program_id(0)
    # FIXME(woosuk): Because is_greedy_ptr is not None at profiling run,
    # re-compilation may happen during runtime when is_greedy_ptr is None.
    if is_greedy_ptr is None:
        is_greedy = True
    else:
        is_greedy = tl.load(is_greedy_ptr + req_idx)
    if not is_greedy:
       ^
ValueError('Cannot bitcast data-type of size 8 to data-type of size 1')
```

Signed-off-by: Mark McLoughlin <markmc@redhat.com>

Author: markmc

vllm/v1/sample/rejection_sampler.py

@@ -375,7 +375,7 @@ def rejection_greedy_sample_kernel(
     if is_greedy_ptr is None:
         is_greedy = True
     else:
-        is_greedy = tl.load(is_greedy_ptr + req_idx)
+        is_greedy = tl.load(is_greedy_ptr + req_idx).to(tl.int1)
     if not is_greedy:
         # Early exit for non-greedy sampling requests.
         return