Fix np.mean with list/tuple containing ArrayBox objects

[tylerflex]

Summary

Fix autograd.numpy.mean() failing when passed a list/tuple containing ArrayBox objects.

import autograd
import autograd.numpy as np

def f(x):
    return np.mean([x, x+2])

g = autograd.grad(f)
g(0.0)  # Previously raised TypeError, now returns 1.0

Previously raised: TypeError: float() argument must be a string or a real number, not 'ArrayBox'

Root cause

When np.mean() receives a list containing ArrayBox objects, numpy's internal _mean implementation calls ret.dtype.type(result) to cast the output.

Since ArrayBox.dtype returns float64, this becomes np.float64(ArrayBox), which calls float() on the ArrayBox and fails.

Fix

• Add a mean wrapper that converts list/tuple inputs to arrays using autograd's
  array() function before calling the primitive
• Update VJP/JVP registrations to use _primitive_mean (the underlying primitive)
• Add **kwargs to grad_np_mean to accept optional arguments like dtype

ran tests and linting with nox