Fix hook memory leak in MRFI by tracking and exposing RemovableHandle references

[Ziad-Ijja]

This pull request fixes a memory leak that occurs when MRFI is instantiated multiple times over a fault injection campaign. The main theme is: making MRFI hooks explicitly managed so they can be properly released.

Root cause:
__add_hooks registered forward pre-hooks and forward hooks via register_forward_pre_hook / register_forward_hook, but the returned RemovableHandle objects were discarded. PyTorch keeps hooks alive as long as their handle exists, meaning every MRFI instantiation accumulated closures (including references to FI_config and internal state) that were never freed, even after the MRFI object itself was deleted. Over a long campaign this causes progressive GPU/CPU memory exhaustion.

Changes in mrfi.py:

• Introduced _hook_handles (list) and _closed (bool) instance attributes, initialized in __init__ before __add_hooks is called.
• Modified __add_hooks to append every RemovableHandle returned by register_forward_pre_hook and register_forward_hook to _hook_handles, instead of discarding them.
• Added remove_hooks(): iterates _hook_handles and calls .remove() on each handle, then clears the list.
• Added close(): calls remove_hooks(), then removes the FI_config attribute from every submodule via module.modules(), and sets _closed = True to prevent double-closing. This fully detaches MRFI instrumentation from the underlying model, allowing the same model instance to be safely re-wrapped in a subsequent MRFI instantiation.

Tests are passing with pytest --cov=mrfi

[codecov-commenter]

Welcome to Codecov 🎉

Once you merge this PR into your default branch, you're all set! Codecov will compare coverage reports and display results in all future pull requests.

Thanks for integrating Codecov - We've got you covered ☂️