[Feature] Add numpy array protocol and enhanced dlpack support to OrtValue

[Copilot]

This PR implements numpy array protocol compatibility and enhanced dlpack support for onnxruntime.OrtValue, enabling seamless integration with the Python tensor ecosystem.

Features Added

1. Numpy Array Protocol Support (__array__)

Enables OrtValue to work seamlessly with numpy functions:

import onnxruntime as ort
import numpy as np

ort_value = ort.OrtValue.ortvalue_from_numpy(np.array([1, 2, 3]))

# Now these work automatically:
numpy_array = np.array(ort_value)
asarray_result = np.asarray(ort_value)

2. Enhanced DLPack Protocol Support

Exposes dlpack methods directly on the Python OrtValue class:

# Direct dlpack protocol access
dlpack_capsule = ort_value.__dlpack__()
device_info = ort_value.__dlpack_device__()
dlpack_capsule2 = ort_value.to_dlpack()  # Convenience method

3. Intelligent from_dlpack() Method

Automatically detects and calls __dlpack__() on source objects:

import torch

# Before: Required manual __dlpack__ call
torch_tensor = torch.tensor([1.0, 2.0])
ort_value = ort.OrtValue.from_dlpack(torch_tensor.__dlpack__())

# After: Automatic detection and calling
ort_value = ort.OrtValue.from_dlpack(torch_tensor)  # Auto-calls __dlpack__!

# Works with any object that has __dlpack__ method
ort_value1 = ort.OrtValue.ortvalue_from_numpy(array)
ort_value2 = ort.OrtValue.from_dlpack(ort_value1)  # Auto-calls __dlpack__

Implementation Details

• Minimal Changes: Only 56 lines added to core implementation
• Full Backward Compatibility: All existing APIs continue to work unchanged
• Proper Delegation: Python methods delegate to existing C++ implementations
• Type Safety: Complete type annotations for better developer experience
• Comprehensive Testing: Added extensive test coverage for all new functionality

Backward Compatibility

All existing functionality is preserved:

• ortvalue._ortvalue.to_dlpack() still works
• OrtValue.from_dlpack(dlpack_capsule, is_bool_tensor=False) still works
• No breaking changes to any existing APIs

Benefits

1. Pythonic Integration: OrtValue now works seamlessly with numpy ecosystem
2. Cross-Library Compatibility: Easy tensor sharing between PyTorch, CuPy, JAX, etc.
3. Simplified API: No need to manually call __dlpack__() methods
4. Modern Standards: Implements current Python tensor protocols

Fixes #24071.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.