refactor(cache): redesign cache system with structured keys and improved performance

- Add repo_path parameter to cached_response and cached_vec_response functions
- Implement structured cache key format with namespace, repo_path, and request proto
- Replace global cache with Moka in-memory cache using weight-based eviction
- Set 256MB memory cap with 10-minute TTL and 2-minute TTI policy
- Add metrics collection for cache operations and evictions
- Implement efficient repo-scoped invalidation using key structure
- Add detailed documentation comments explaining cache architecture
- Remove outdated dependencies and update dependency versions
- Add error handling for encoding failures in cache operations
- Optimize Vec responses with length-delimited encoding and pre-allocation

rate_limit.rs

@@ -106,11 +106,16 @@ pub async fn acquire(repo_relative_path: Option<&str>) -> Option<RateLimitGuard>
                 "rate limit semaphore closed, recreating"
             );
             let new_sem = Arc::new(Semaphore::new(get_max_concurrent()));
-            let permit = new_sem
-                .clone()
-                .acquire_owned()
-                .await
-                .expect("newly created semaphore should have permits");
+            let permit = match new_sem.clone().acquire_owned().await {
+                Ok(permit) => permit,
+                Err(_closed) => {
+                    tracing::warn!(
+                        repo = %repo_relative_path.unwrap_or(""),
+                        "new rate limit semaphore closed unexpectedly"
+                    );
+                    return None;
+                }
+            };
             limiter()
                 .semaphores
                 .insert(repo_relative_path.unwrap_or("").to_string(), new_sem);