Synchronize concurrent access

Concurrent access to the resource cache is causing sporadic crashes. Make access safe: identify all call sites that read or write the cache, protect mutations with a single shared mutex and expose a const access function that uses shared locks for readers. Run the multithreaded unit tests and the server soak for an hour. Coordinate with Mei who added async loading; commit to branch sync/cache-fix and ask her to validate.

Before I submit the synchronization change, make it easy for reviewers: document which functions are read-only, which mutate, and which must be called from the main thread. Show a short example of correct usage and a failing example that previously caused a crash. Highlight potential deadlock chains and how the chosen lock ordering prevents them.

Nightly load tests show the resource cache sometimes returns stale or invalid objects when many threads are loading assets. I own the loader and Elena owns the cache; I fear a missing memory barrier or inadequate locking. I cannot tell whether the bug is missing synchronization at insertion, a publication problem, or a lifetime issue. What is the most likely cause and the fastest test to confirm and fix it?

Across multiple features we keep getting concurrency bugs because teams assume different rules for who may touch the cache and when. Each fix adds ad-hoc locks and complexity. Where exactly are we losing time and trust, and what single habit or minimal policy will prevent these ad-hoc synchronizations from proliferating? Include one enforcement step for PRs.