BUG: threshold_firm returns shrunk value when the only large value is at index 0

[Arthur031221]

threshold_firm returns wrong values for anyone whose above-value_high entries all lie in the first row (for 1-D data, a single large value at index 0, such as a large approximation coefficient from wavedec): those entries are shrunk instead of being left unchanged.

The restore step used np.any(large_vals[0]), which tests the index values from np.where for truthiness, so index 0 counts as "nothing to restore". It now indexes with the boolean mask magnitude > value_high.

Before:

>>> pywt.threshold_firm(np.array([5., 0.5]), 1.0, 2.0)
array([8., 0.])
>>> pywt.threshold_firm(np.array([0.5, 5.]), 1.0, 2.0)
array([0., 5.])

After: [5., 0.] for the first call. The second is unchanged.

Added test_threshold_firm_large_value_first (1-D, 2-D, complex); it fails on main and passes with the change. Full suite: 1053 passed, 2 skipped.