Return zero energy for a module with no Vodes

[cemde]

Bug

return functools.reduce(lambda x, y: x + y, (m.energy() for m in self.submodules(cls=EnergyModule)))

functools.reduce with no initial value raises on an empty iterable rather than returning an identity.

Impact

A module with no EnergyModule children raises TypeError: reduce() of empty iterable with no initial value instead of returning zero. That hits any model built up incrementally, and any leaf module in a hierarchy that happens to hold no Vode.

Fix

Seed the reduction with jax.numpy.asarray(0.0).

Why that value

Four candidates, measured:

• 0 breaks jax.grad on the empty case: grad requires real-valued outputs, got int32.
• 0.0 works numerically but returns a Python float for the empty case, contradicting the method's documented jax.Array return, and costs one extra ty diagnostic.
• jnp.zeros(()) is strongly typed float32, so it silently upcasts a bfloat16 or float16 energy in the non-empty case.
• jax.numpy.asarray(0.0) is a jax.Array that keeps weak_type=True, so it adopts the operand's dtype and leaves the non-empty result untouched.

Confirmed bit-identical to main on the non-empty path, both unbatched and under pxf.vmap.

Closes #72

[liukidar]

Lgtm