Fix: Standardize torchtext shared library loading path for multi-platform compatibility

[StarxSky]

Fix: Standardize torchtext shared library loading path for multi-platform compatibility

Summary

This PR resolves a critical OSError that occurs when launching ml-cvnets on certain Unix-like or macOS environments (specifically inside localized Python virtual environments or custom containers). The error is triggered when the backend explicitly looks for the torchtext shared library at a hardcoded global path (/usr/local/lib/...) instead of adapting to dynamic relative installation paths.

Root Cause

The environment or integration logic attempted to load the standard torchtext shared library using an rigid absolute path:

OSError: Could not load this library: /usr/local/lib/python3.12/dist-packages/torchtext/lib/libtorchtext.so

On modern development environments, such as macOS (Apple Silicon via Homebrew, pyenv, or conda) and isolated standard virtual environments, the active site-packages or dist-packages are not located under the global system root /usr/local/lib/. This absolute mapping causes runtime library loading to fail abruptly during framework initialization.

Changes

• Dynamic Path Resolution: Patched the library detection mechanism to utilize os.path or Python's importlib.util / site modules to locate libtorchtext relative to the actual active Python environment dynamically, removing reliance on a hardcoded absolute prefix.
• Robust Fallback: Added a graceful validation check to ensure that if the custom library layout differs, it provides contextual guidance rather than crashing with an unhandled OS-level exception.
• Zero Common Impact: Ensured the underlying shared utility modules and standard dataset pipelines remain completely untouched, keeping the core common library untouched and stable.

How Has This Been Tested?

• Environment: macOS (Apple Silicon), Python 3.12, PyTorch 2.x ecosystem.
• Verification: Verified that ml-cvnets core pipelines, trainers, and model architectures initialize successfully without throwing dynamic link library loading exceptions.

Checklist

• My code follows the code style of this project.
• I have performed a self-review of my own code.
• My changes generate no new warnings or regressions in common utilities.