Constrained optimization using `projection_polyhedron` claims the polyhedron is empty when it is not.

A minimal example:

```.py
import jaxopt
import jax.numpy as jnp

def objective(x):
return -jnp.sum(jnp.dot(jnp.array([2, 1], jnp.float32), x))

def main():
a = jnp.array([[0, 0]], jnp.float32)
b = jnp.array([0], jnp.float32)
g = jnp.array([[0, 0], [0, 0]], jnp.float32)
h = jnp.array([0, 0], jnp.float32)

pg = jaxopt.ProjectedGradient(dummy, jaxopt.projection.projection_polyhedron, jit=False)
result = pg.run(jnp.array([0, 0], jnp.float32), hyperparams_proj=(a, b, g, h))
return result
```

The equality and inequality conditions are tautological, but running `main` still results in a raised error:

```py
ValueError: The polyhedron is empty.
```

Other variations of equality and inequality constraints that clearly correspond to non-empty polyhedrons likewise raise the same error. Any guidance or advice would be appreciated. Thank you.