[bug] TracerWarning from subsample_messenger when running NUTS with jit_compile=True #3449

/opt/miniconda3/envs/pyro/lib/python3.10/site-packages/pyro/poutine/subsample_messenger.py:70:
TracerWarning: torch.tensor results are registered as constants in the trace.
You can safely ignore this warning if you use this function to create tensors out of constant variables that would be
the same every time you call this function.
In any other case, this might cause the trace to be incorrect.
  result = torch.tensor(0.0, device=self.device)

This appears even when the model runs successfully and the warning seems to come from Pyro internals rather than user
model code.

I understand this may be harmless because the tensor is a constant, but it is confusing for users because:

1. the warning looks like a potential tracing bug,
2. it points to Pyro internal code instead of user code,
3. the only obvious way to suppress it is to disable jit_compile=True, which may reduce performance.

I am opening this issue to ask whether this warning is expected under JIT-compiled NUTS, and whether Pyro should
suppress or avoid triggering it internally.

### Environment

- OS: macOS 26.3.1 (arm64)
- Python: 3.10.19
- PyTorch: 2.5.1
- Pyro: 1.9.1+ab0491a

### Code Snippet

A minimal pattern that triggers the warning is:

import pyro
from pyro.infer import MCMC, NUTS

nuts = NUTS(model, jit_compile=True)
mcmc = MCMC(nuts, warmup_steps=1000, num_samples=1000, num_chains=4)
mcmc.run(obs=data)

In my case the warning is raised from:

result = torch.tensor(0.0, device=self.device)

inside pyro/poutine/subsample_messenger.py.