feat: add panic tracking and logging with PanicTracker

- Implemented PanicTracker to log panics using tracing within a dedicated span.
- Introduced PanicInfo struct to capture panic details.
- Added tests for panic catching and hook restoration.

feat: implement token-bucket rate limiter

- Created RateLimiter to manage token consumption with configurable rates and burst capacity.
- Added methods for acquiring tokens and checking available tokens.
- Included tests for rate limiting behavior and token availability.

feat: add automatic retry with exponential backoff

- Developed retry function to handle transient errors with configurable retry strategies.
- Introduced RetryConfig for customizing retry behavior and jitter.
- Added tests for various retry scenarios and error handling.

feat: implement graceful shutdown coordination

- Created ShutdownManager to manage shutdown signals and task completion.
- Introduced ShutdownSignal for tasks to observe shutdown requests.
- Added tests for shutdown request handling and task draining.

feat: add timeout functionality for async operations

- Implemented with_timeout function to enforce execution time limits on futures.
- Created Timeout struct for wrapping futures with a deadline.
- Added tests for timeout behavior and error mapping.

crates/corex-cache/src/multilayer.rs

@@ -0,0 +1,139 @@
+//! Multi-layer (L1/L2) caching behind a single [`Cache`] face.
+//!
+//! [`MultiLayerCache`] layers a fast in-process L1 over a slower but larger
+//! L2 (e.g. Redis). Reads are L1-first with an L2 fallback; a hit on L2 is
+//! backfilled into L1. Writes and invalidations go to both layers.
+
+use std::time::Duration;
+
+use async_trait::async_trait;
+
+use crate::traits::{Cache, CacheError};
+
+/// A read-through, write-through composition of an L1 and L2 cache.
+///
+/// `L1` is typically [`crate::memory::MemoryCache`] or
+/// [`crate::moka_cache::MokaL1`]; `L2` is typically a distributed cache such
+/// as a Redis backend. Order of layers fixed: `L1` is consulted first.
+#[derive(Clone)]
+pub struct MultiLayerCache<L1, L2> {
+    l1: L1,
+    l2: L2,
+    /// When `true`, an L2 hit is written back into L1 (default `true`).
+    populate_l1: bool,
+}
+
+impl<L1, L2> MultiLayerCache<L1, L2>
+where
+    L1: Cache,
+    L2: Cache,
+{
+    /// Builds a two-layer cache with L1-backfill enabled.
+    pub fn new(l1: L1, l2: L2) -> Self {
+        Self {
+            l1,
+            l2,
+            populate_l1: true,
+        }
+    }
+
+    /// Disables L1 backfill-on-read.
+    pub fn without_l1_backfill(mut self) -> Self {
+        self.populate_l1 = false;
+        self
+    }
+
+    /// Returns a reference to the L1 layer.
+    pub fn l1(&self) -> &L1 {
+        &self.l1
+    }
+
+    /// Returns a reference to the L2 layer.
+    pub fn l2(&self) -> &L2 {
+        &self.l2
+    }
+}
+
+#[async_trait]
+impl<L1, L2> Cache for MultiLayerCache<L1, L2>
+where
+    L1: Cache,
+    L2: Cache,
+{
+    async fn get(&self, key: &str) -> Result<Option<Vec<u8>>, CacheError> {
+        // L1 first.
+        if let Some(value) = self.l1.get(key).await? {
+            return Ok(Some(value));
+        }
+        // L2 fallback.
+        if let Some(value) = self.l2.get(key).await? {
+            if self.populate_l1 {
+                self.l1.set(key, value.clone(), None).await?;
+            }
+            return Ok(Some(value));
+        }
+        Ok(None)
+    }
+
+    async fn set(
+        &self,
+        key: &str,
+        value: Vec<u8>,
+        ttl: Option<Duration>,
+    ) -> Result<(), CacheError> {
+        self.l1.set(key, value.clone(), ttl).await?;
+        self.l2.set(key, value, ttl).await
+    }
+
+    async fn invalidate(&self, key: &str) -> Result<(), CacheError> {
+        self.l1.invalidate(key).await?;
+        self.l2.invalidate(key).await
+    }
+
+    async fn clear(&self) -> Result<(), CacheError> {
+        self.l1.clear().await?;
+        self.l2.clear().await
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::memory::MemoryCache;
+
+    #[tokio::test]
+    async fn read_through_populates_l1() {
+        let l2 = MemoryCache::new();
+        l2.set("k", b"l2-value".to_vec(), None).await.unwrap();
+
+        let layered = MultiLayerCache::new(MemoryCache::new(), l2);
+        assert_eq!(layered.l1().get("k").await.unwrap(), None);
+        assert_eq!(layered.get("k").await.unwrap(), Some(b"l2-value".to_vec()));
+        // L2 hit should have populated L1.
+        assert_eq!(
+            layered.l1().get("k").await.unwrap(),
+            Some(b"l2-value".to_vec())
+        );
+    }
+
+    #[tokio::test]
+    async fn write_goes_to_both() {
+        let l1 = MemoryCache::new();
+        let l2 = MemoryCache::new();
+        let layered = MultiLayerCache::new(l1, l2.clone());
+        layered.set("k", b"v".to_vec(), None).await.unwrap();
+        assert_eq!(layered.l1().get("k").await.unwrap(), Some(b"v".to_vec()));
+        assert_eq!(l2.get("k").await.unwrap(), Some(b"v".to_vec()));
+    }
+
+    #[tokio::test]
+    async fn invalidate_clears_both() {
+        let l1 = MemoryCache::new();
+        let l2 = MemoryCache::new();
+        let layered = MultiLayerCache::new(l1, l2);
+        layered.set("k", b"v".to_vec(), None).await.unwrap();
+        layered.invalidate("k").await.unwrap();
+        assert_eq!(layered.l1().get("k").await.unwrap(), None);
+        assert_eq!(layered.l2().get("k").await.unwrap(), None);
+    }
+}