Wrap cache reader in BufReader for speed. (#166)

Profiling reading from the cache in the download of a 1GB file showed
that 48% of the time was spent in system read calls. This PR removes
that bottleneck by simply wrapping the read calls in a BufReader. This
eliminates that bottleneck and showed a 15-20% overall speed improvement
restoring the 1GB file from cache (based on a few non-rigorous bash
`time` runs).

chunk_cache/src/disk.rs

@@ -264,15 +264,17 @@ impl DiskCache {
                 }
             }
 
-            let Ok(header) = CacheFileHeader::deserialize(&mut file)
+            let mut file_reader = std::io::BufReader::new(file);
+
+            let Ok(header) = CacheFileHeader::deserialize(&mut file_reader)
                 .debug_error(format!("failed to deserialize cache file header on path: {path:?}"))
             else {
                 self.remove_item(key, &cache_item)?;
                 continue;
             };
 
             let start = cache_item.range.start;
-            let result_buf = get_range_from_cache_file(&header, &mut file, range, start)?;
+            let result_buf = get_range_from_cache_file(&header, &mut file_reader, range, start)?;
             return Ok(Some(result_buf));
         }
     }