Support Lineax operators for the vector field of ControlTerm

By providing support for lineax.AbstractLinearOperators in the vector field of a ControlTerm it may be possible to reduce the need for WeaklyDiagonalControlTerm and/or any other such specialised control terms.

The following gives a MWE:

class ControlTerm(_ControlTerm):
    def prod(self, vf, control):
        if isinstance(vf, lx.AbstractLinearOperator):
            return jtu.tree_map(lambda _vf, _control: _vf.mv(_control), vf, control)
        return jtu.tree_map(_prod, vf, control)

# These two are now equivalent.
ControlTerm(lx.DiagonalLinearOperator(jnp.array([1,2,3])), ...)
WeaklyDiagonalControlTerm(jnp.array([1,2,3]), ...)

Not sure if this is something you want to support, but it occurred to me that the operator tags might also be useful for some of the diffrax solvers?

Start at the ControlTerm vector field and its prod entry point, comparing the existing ControlTerm and WeaklyDiagonalControlTerm behavior shown in the MWE. Check how lineax.AbstractLinearOperators and their mv operation should interact with the control tree, then verify that the two example constructions are equivalent without regressing the existing behavior.