fix: avoid panic when the file is corrupted in disk cache (#1130)

## Rationale
- The corrupted file in the disk cache will lead to the panic of the
server
- The whole cache page file is fetched when only a range of the file is
accessed, leading to high memory consumption

## Detailed Changes
- Store the file size in the memory for file integrity check
- Omit the corrupted file in the disk cache rather than panic
- Do IO operation outside the meta data cache lock
- Read only the involved range of a cached paged file instead of the
whole file

## Test Plan
Existing tests and a newly designed test.

analytic_engine/src/setup.rs

@@ -519,12 +519,18 @@ fn open_storage(
             }
         };
 
+        store = Arc::new(StoreWithMetrics::new(
+            store,
+            engine_runtimes.io_runtime.clone(),
+        ));
+
         if opts.disk_cache_capacity.as_byte() > 0 {
             let path = Path::new(&opts.disk_cache_dir).join(DISK_CACHE_DIR_NAME);
             tokio::fs::create_dir_all(&path).await.context(CreateDir {
                 path: path.to_string_lossy().into_owned(),
             })?;
 
+            // TODO: Consider the readonly cache.
             store = Arc::new(
                 DiskCacheStore::try_new(
                     path.to_string_lossy().into_owned(),
@@ -538,11 +544,6 @@ fn open_storage(
             ) as _;
         }
 
-        store = Arc::new(StoreWithMetrics::new(
-            store,
-            engine_runtimes.io_runtime.clone(),
-        ));
-
         if opts.mem_cache_capacity.as_byte() > 0 {
             let mem_cache = Arc::new(
                 MemCache::try_new(