Implement new MomentumQNGOptimizer compatible with (q)jit and program capture

[SimoneGasperini]

The goal here is to implement a new version of the MomentumQNGOptimizer based on Jax and compatible with jax.jit, qml.qjit, and program capture. This PR depends on the q(jit)/capture compatible version of the standard QNGOptimizer, see #7452 for more context.

Note that this optimization algorithm requires to keep track of a non-trivial internal state: in the current Autograd-based implementation, this state is stored in the optimizer's accumulation attribute as a pennylane.numpy array (see here). However, in this Jax-based version, the state needs to be carried over at each optimization step as it happens for the native optax optimizers. This is required to avoid having a Jax tracer as a class attribute when using (q)jit compilation.

[sc-92558]

[github-actions]

Hello. You may have forgotten to update the changelog!
Please edit doc/releases/changelog-dev.md with:

• A one-to-two sentence description of the change. You may include a small working example for new features.
• A link back to this PR.
• Your name (or GitHub username) in the contributors section.