fix(caching): make cache_disabled context-local using ContextVar

[Aryan-Pardeshi]

Fixes #1983

Context

\ cache_disabled()\ directly mutated the global _caching_enabled\ boolean. In asynchronous environments, calling \with cache_disabled():\ in one task inadvertently disabled caching globally for concurrent tasks running in the same process loop.

Changes

• Added a \ContextVar[Optional[bool]]\ (_caching_enabled_override) defaulting to \None.
• Implemented _is_caching_enabled()\ to check the context-local override first, falling back to global _caching_enabled\ if \None.
• Updated \cache_disabled()\ context manager to set and reset the \ContextVar\ token.
• Retained global _caching_enabled\ so process-level switches like \disable_cache()\ continue to apply across worker threads.
• Added unit tests in \ ests/test_cache.py\ covering async task isolation and thread behavior.

[harsh4vardhan]

The ContextVar approach is exactly right for this, and the two-level design (_caching_enabled_override + fallback to global _caching_enabled) cleanly preserves the intended semantics: disable_cache() remains process-wide, cache_disabled() becomes context-local. Token-based set/reset handles nesting correctly too.

One thing worth documenting: asyncio.create_task() called inside a cache_disabled() block will spawn a task that inherits _caching_enabled_override = False - standard ContextVar propagation means the child sees the parent's context at spawn time. That may be the desired behavior (subtasks honor the same override), but it is potentially surprising. A sentence in the docstring would help users reason about it.

The async decorator path still carries # pragma: no cover. The new async tests hit _is_caching_enabled() correctly, but the if not _is_caching_enabled() guard inside the async wrapper itself stays excluded from the coverage report. Not a blocker, but you may want to remove the pragma or add a direct test through @cache on an async def function.