Hi there, I just started using numpyro to build a hierarchical model where a group-level parameter is sampled from a circular distribution. I use `handler.reparam`

to set this parameter as `CircularReparam()`

.

```
def hierarchical_model(sub_idx, xx, yy=None):
# Define hyperpriors mu_m, sigma_m for m
mu_m = numpyro.sample('mu_m', dist.Uniform(-numpy.pi, numpy.pi))
sigma_m = numpyro.sample('sigma_m', dist.HalfNormal(50.))
# Sample m
with numpyro.plate("plate_i", n_sub):
with handlers.reparam(config={'m': CircularReparam()}):
m = numpyro.sample('m', dist.VonMises(mu_m, sigma_m))
# Sample estimates
.....
```

The model can run without error, but I cannot read the data with `az.from_numpyro()`

because of a NotImplementedError in `~/numpyro/distributions/distribution.py`

, line 249. The error is gone when I remove the reparam config for `m`

.

Do you maybe have suggestions to solve the problem?