[MPS] Raise NotImplementedError for pooling ops that require 64-bit indexing (#194460)

Continuation of #194373
More errors. Below script causes failures:
```
import torch
import torch.nn.functional as F

def run(name, shape, op, bwd):
    x = torch.randn(shape, dtype=torch.half)
    xm = x.to("mps").requires_grad_(bwd)
    xc = x.float().requires_grad_(bwd)
    try:
        ym, yc = op(xm), op(xc)
        if bwd:
            g = torch.randn_like(yc)
            ym.backward(g.to("mps").half())
            yc.backward(g)
            rm, rc = xm.grad.cpu().float(), xc.grad
        else:
            rm, rc = ym.detach().cpu().float(), yc.detach()
        bad = int(((rm - rc).abs() > 0.1).sum())
        print(f"{name}: {'FAIL' if bad else 'ok'}  mismatched={bad}")
    except NotImplementedError as e:
        print(f"{name}: guarded ({e})")
    torch.mps.empty_cache()

P2 = (34, 1, 8192, 8192)
P3 = (70, 32, 100, 100, 100)

run("max/adaptive fwd template", P2, lambda t: F.max_pool2d(t, 2), False)
run("max bwd template", P2, lambda t: F.adaptive_max_pool2d(t, 128), True)
run("avg fwd template", P2, lambda t: F.avg_pool2d(t, 2, ceil_mode=True), False)
run("avg bwd template", P3, lambda t: F.avg_pool3d(t, 2), True)
```

leads to:
```
max/adaptive fwd template: FAIL  mismatched=32208577
max bwd template: FAIL  mismatched=30132
avg fwd template: FAIL  mismatched=28232769
avg bwd template: FAIL  mismatched=39205174
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/194460
Approved by: https://github.com/malfet