BUG: clip_evals uses wrong indicator

[josef-pkt]

def clip_evals(x, value=0): #threshold=0, value=0):
    evals, evecs = np.linalg.eigh(x)
    clipped = np.any(evals < 0)
    x_new = np.dot(evecs * np.maximum(evals, value), evecs.T)
    return x_new, clipped

we are clipping based on value and not necessarily at 0
clipped = np.any(evals < value)

this is used as indicator in the functions to see whether clipping was necessary. So, that's incorrect if threshold (value) != 0.
(in examples, value was still tiny like 1e-15, IIRC, just to get positive definiteness. So there is a small imprecision in the results because of this bug)