Make Tensor.__dlpack__(stream=None) capture-safe during CUDA Graph capture (pytorch#163242)

Many extensions (including pybind helpers) call `Tensor.__dlpack__()` without a stream argument. Before pytorch#150217, `stream=None` behaved like “no cross-stream sync” and was safe inside CUDA Graph capture. After pytorch#150217, `stream=None` maps to the legacy default stream, adding a cross-stream wait that invalidates capture when running on a non-default stream.

See this example

```
import torch
s = torch.cuda.Stream()
x = torch.randn(8, device="cuda")
g = torch.cuda.CUDAGraph()

with torch.cuda.stream(s):
    with torch.cuda.graph(g):
        _ = x + 1
        cap = x.__dlpack__()
        _ = torch.utils.dlpack.from_dlpack(cap)
```

This PR partially reverts pytorch#150217 that stream=None defaults to no sync.

Pull Request resolved: pytorch#163242
Approved by: https://github.com/ngimel

Author: eee4017

torch/_tensor.py

@@ -1699,7 +1699,7 @@ def __torch_function__(cls, func, types, args=(), kwargs=None):
     def __dlpack__(
         self,
         *,
-        stream: Optional[Any] = None,
+        stream: Optional[Any] = -1,
         max_version: Optional[tuple[int, int]] = None,
         dl_device: Optional[tuple[enum.IntEnum, int]] = None,
         copy: Optional[bool] = None,
@@ -1717,9 +1717,12 @@ def __dlpack__(
                 pointer to a CUDA stream. The current stream is synchronized with
                 this stream before the capsule is created, and since the capsule
                 shares its storage with the tensor this make it safe to access from
-                both streams.  If None or -1 is passed then no synchronization is performed.
+                both streams.  If -1 is passed then no synchronization is performed.
                 If 1 (on CUDA) or 0 (on ROCM) then the default stream is used for
-                synchronization.
+                synchronization. This API intentionally slightly deviates from the DLPack
+                guidance: the default stream is -1 (stream-preserving; no cross-stream sync),
+                because many from_dlpack implementations intend stream preservation.
+                For non-CUDA devices, -1 is treated the same as None.
 
             max_version (tuple[int, int] or None): An optional Python tuple with
                 2 integers, representing the maximum version the caller supports. If
@@ -1797,7 +1800,7 @@ def __dlpack__(
                 event.record(current_stream)
                 stream.wait_event(event)
         elif self.device.type == "cpu":
-            assert stream is None, "stream should be None on cpu."
+            assert stream is None or stream == -1, "stream should be None on cpu."
 
         if self.device.type == "xla":
             import torch_xla