`torch.argsort` is not stable by default

[mdhaber]

According to the standard for argsort, the default should be stable=True. So I'd expect these two argsorts to produce the same result (xp.arange(30), essentially), but they do not.

# array_api_compat version 1.12.0
from array_api_compat import torch as xp
x = xp.asarray([10]*30)

xp.argsort(x)
# tensor([15, 29, 28, 27, 26, 25, 24, 23, 22, 21, 20, 19, 18, 17, 16,  0, 14, 13,
#         12, 11, 10,  9,  8,  7,  6,  5,  4,  3,  2,  1])

xp.argsort(x, stable=True)
# tensor([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16, 17,
#         18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29])

[cakedev0]

I can see why when reading the code, I'll open a PR.