[Dynamo] Constant-fold CUDA Tensor.is_pinned() (#186494)

Dynamo traced Tensor.is_pinned() as a generic tensor method call. In user
Triton paths such as Triton's pinned-memory helper, this meant a CUDA fake
tensor produced an aten.is_pinned bool result while tracing. Older builds
failed while fake-running aten.is_pinned; current main has FakeTensor coverage
but still cannot wrap the bool-producing torch op in the FX graph.

For non-CPU tensors, ATen is_pinned() is metadata-independent and always
false. Add a TensorVariable method handler that returns a Python constant
False for the no-argument non-CPU case before creating an FX proxy. Leave CPU
tensors on the generic path because CPU pinned memory can vary independently
of Dynamo's tensor metadata guards.

This keeps the fix limited to the sound case needed by CUDA user Triton code
instead of treating all is_pinned calls as metadata constants.

Fixes #121526
Generated by my agent

Test Plan:
- python test/dynamo/test_misc.py -k test_cuda_tensor_is_pinned_constant_false
- lintrunner -a
- git diff --check

Pull Request resolved: https://github.com/pytorch/pytorch/pull/186494
Approved by: https://github.com/eellison, https://github.com/mlazos