initial_chunk_size_bytes is a signed int (limited to 2 GiB)

Nobody has claimed this yet.

### Describe the issue

For use with multiple sessions and a shared memory arena, where I know that each session will use no more than X GiB, [the documentation](https://onnxruntime.ai/docs/get-started/with-c.html) suggests setting `initial_chunk_size_bytes` such that most model requests will be completed without needing to allocate more memory.

Unfortunately, `initial_chunk_size_bytes` is a signed int ([struct definition here](https://github.com/microsoft/onnxruntime/blob/6ed77cc374c04956bc1197d0dc0fe5a2ed9b15b9/include/onnxruntime/core/framework/allocator.h#L17-L40)), and thus has a maximum value of `2**31 - 1` or just under 2 GiB. In my case, I know that my models will never use more than 4 GiB during inference, but I cannot set this, as it is too large.

Is there a reason not to use `int64_t` for this?

### To reproduce

This is reproducible in either C++ or Python. Here's the Python:
```python3
import onnxruntime

# This works:
arena_cfg = onnxruntime.OrtArenaCfg({'initial_chunk_size_bytes': 2 * 1024 * 1024 * 1024 - 1})

# This throws an exception:
arena_cfg = onnxruntime.OrtArenaCfg({'initial_chunk_size_bytes': 2 * 1024 * 1024 * 1024})
```

### Urgency

_No response_

### Platform

Linux

### OS Version

Ubuntu 22

### ONNX Runtime Installation

Released Package

### ONNX Runtime Version or Commit ID

1.17.0

### ONNX Runtime API

Python

### Architecture

X64

### Execution Provider

CUDA

### Execution Provider Library Version

_No response_

The allocator configuration struct is linked in include/onnxruntime/core/framework/allocator.h; start by inspecting initial_chunk_size_bytes and tracing the Python OrtArenaCfg binding. Reproduce the 2 GiB boundary from the issue, then verify that a 4 GiB value is accepted without the reported exception in the affected APIs.

Written by the indexing model from the issue text.