`verbose=False` is not working as expected for `NonlinearCG`

[Joshuaalbert]

    def loss(U: UType):
        return -model.log_prob_likelihood(U, allow_nan=False)

    solver = NonlinearCG(
        fun=loss,
        jit=True,
        unroll=False,
        verbose=False
    )
    results = solver.run(init_params=init_U_point)

It still prints out linesearch debug lines like:

WARNING: jaxopt.ZoomLineSearch: Returning stepsize with sufficient decrease but curvature condition not satisfied.
INFO: jaxopt.ZoomLineSearch: Iter: 9 Minimum Decrease & Curvature Errors (stop. crit.): 3.3350500139306405e-09 Stepsize:0.00016820144082885236  Decrease Error:0.0  Curvature Error:3.3350500139306405e-09 

[vroulet]

Thanks for bringing this up. https://github.com/google/jaxopt/pull/573 should fix the issue, but I need to investigate the failing tests.