cumsum and cumprod modify the input array

While debugging a test failure in cupyimg, I found that cupy.cumsum unexpectedly modifies the input array. This did not used to be the case, so I suspect it may be due to recent changes from #4316 or #4366 (attn: @anaruse)

• Conditions (you can just paste the output of python -c 'import cupy; cupy.show_config()') CuPy 9.0.0a3

• Code to reproduce

Here is a basic example to illustrate the issue:

a = cp.testing.shaped_arange((4, 4))
print(a)
b = np.cumsum(a, axis=1)
print(a)  # a will now be equal to b (it was modified in-place)

array([[ 1., 2., 3., 4.], [ 5., 6., 7., 8.], [ 9., 10., 11., 12.], [13., 14., 15., 16.]], dtype=float32)

array([[ 1., 3., 6., 10.], [ 5., 11., 18., 26.], [ 9., 19., 30., 42.], [13., 27., 42., 58.]], dtype=float32)

[grlee77]

My environment was set to CUPY_ACCELERATORS=cub,cutensor when I encountered this

[leofang]

I guess the bug is this line (from #4316): https://github.com/cupy/cupy/blob/3388dcec8c939eb01baecb2a0459e4116db05b1d/cupy/core/_routines_math.pyx#L742 copy should be True.