[BUG] Fix thread-unsafe singleton initialization in factory functions

[direkkakkar319-ops]

What does this implement/fix? Explain your changes.

All five singleton factory functions (get_registry, get_handle_manager, get_executor, get_tag_resolver, get_composition_validator) used a simple if None check with no locking, making them unsafe under concurrent access. When fit_predict_async offloads work to thread pool executors via loop.run_in_executor(), multiple threads can race through the if _instance is None check simultaneously, creating duplicate instances and leading to inconsistent state (e.g., duplicate registry loads, split handle stores).
This PR applies the double-checked locking pattern to all five singletons using threading.Lock

if _instance is None:          # Fast path
    with _lock:                # Serialize only during first initialization
        if _instance is None:  # Re-check under lock
            _instance = Class()

• Each file receives two minimal changes:

1. import threading added to imports
2. A module-level threading.Lock() and the double-checked locking pattern in the factory function

Does your contribution introduce a new dependency? If yes, which one?

No. threading is part of the Python standard library.

What should a reviewer concentrate their feedback on?

1. Correctness of the double-checked locking pattern across all five files
2. Whether any of these singletons should additionally protect their internal mutable state with locks for thread-safe mutation, beyond just safe initialization .

Any other comments?

Double-checked locking was used instead of functools.lru_cache(maxsize=1) because it keeps the existing global pattern (so reset/tests stay simple), avoids cache-related overhead and GC complications, and ensures the fast path has no lock overhead—just a quick None check.

PR checklist

For all contributions

• I've added myself to the list of contributors.
• Optionally, I've updated sktime's CODEOWNERS to receive notifications about future changes to these files.
• I've added unit tests and made sure they pass locally.

For new estimators

None

pytest pass locally

lint pass locally

[direkkakkar319-ops]

Hey! @Shashankss1205 , This PR fixes a race condition in all five singleton factory functions. Since fit_predict_async uses loop.run_in_executor() to offload work to threads, the existing if None checks could be hit concurrently, creating duplicate instances.