[BUG] initPrx busy-spins on lock contention burning CPU

- severity: Medium
- files: `shenyu-plugin/shenyu-plugin-proxy/shenyu-plugin-rpc/shenyu-plugin-tars/src/main/java/org/apache/shenyu/plugin/tars/cache/ApplicationConfigCache.java:189-206`
- description: `initPrx` is a `while(true)` loop that calls `tryLockedLoadMetaData(...)` whenever `prxClassCache.get(path)` is null. `tryLockedLoadMetaData` uses `LOCK.tryLock()`; if the lock is held, it returns immediately, leaving `prxClass` null, so the loop re-enters with no `Thread.yield()`/backoff.
- impact: Under concurrent tars metadata registration, worker threads spin-wait (100% CPU) for the duration of ByteBuddy class generation.
- suggested_fix: Acquire the lock blockingly inside the loop, or back off with `Thread.yield()` when `tryLock()` fails.
- confidence: Medium
- related_existing: none

---
_Identified during the 2026-08-02 deep re-scan; full list in [`docs/scan2-2026-08-02/06-medium-tiers.md`](docs/scan2-2026-08-02/06-medium-tiers.md)._

Start in shenyu-plugin/shenyu-plugin-proxy/shenyu-plugin-rpc/shenyu-plugin-tars/src/main/java/org/apache/shenyu/plugin/tars/cache/ApplicationConfigCache.java:189-206. Read the initPrx loop together with tryLockedLoadMetaData and observe lock-contention behavior. Done means contention no longer causes a tight CPU-burning loop while TARS metadata and ByteBuddy class generation still complete correctly.

Written by the indexing model from the issue text.