Fix NestedTensor amin/amax/argmin operations for integer dtypes

[lchen2331]

Fix issues: #2703
The issue is caused by an invalid type conversion when the input tensor values are integers while the padding value is of double-float type, resulting in incorrect casting behavior. This is fixed by clamping the padding value to the valid min/max range of the target input dtype.

[Copilot]

Pull request overview

This PR fixes integer type support in amin/amax/argmin operations by addressing invalid type conversions between double-float padding values and integer tensor values. The fix clamps padding values to the valid min/max range of the target dtype.

Changes:

• Replaced direct static_cast of padding_value with a call to _get_padding_value helper function that properly handles type conversion for both floating-point and integer types

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[CuiYifeng]

LGTM.

[CuiYifeng]

Please update PR title since the fixing is for NestedTensor, thanks

[guangyey]

Nice!

[CuiYifeng]

New Passing Known Issues in #2703:

op_ut,third_party.torch-xpu-ops.test.xpu.test_nestedtensor_xpu.TestNestedTensorDeviceTypeXPU,test_jagged_amin_dtypes_xpu_int32
op_ut,third_party.torch-xpu-ops.test.xpu.test_nestedtensor_xpu.TestNestedTensorDeviceTypeXPU,test_jagged_amin_dtypes_xpu_int64
op_ut,third_party.torch-xpu-ops.test.xpu.test_nestedtensor_xpu.TestNestedTensorDeviceTypeXPU,test_jagged_argmin_dtypes_xpu_int32
op_ut,third_party.torch-xpu-ops.test.xpu.test_nestedtensor_xpu.TestNestedTensorDeviceTypeXPU,test_jagged_argmin_dtypes_xpu_int64
op_ut,third_party.torch-xpu-ops.test.xpu.test_nestedtensor_xpu.TestNestedTensorDeviceTypeXPU,test_jagged_min_dtypes_xpu_int32
op_ut,third_party.torch-xpu-ops.test.xpu.test_nestedtensor_xpu.TestNestedTensorDeviceTypeXPU,test_jagged_min_dtypes_xpu_int64

[github-actions]

Performance outliers, please check!

• 🟡 [80%, 90%), may be fluctuations

Category	Model	Target vs. Baseline [Eager]	Target vs. Baseline [Inductor]
torchbench_bfloat16_training	resnet18	0.937951	0.83756