fix: prevent panics from byte-level string slicing on multi-byte UTF-8

Uses floor_char_boundary() instead of direct byte indexing to prevent panics when slicing strings containing multi-byte UTF-8 characters.

src/memory/hygiene.rs

@@ -326,7 +326,7 @@ fn date_prefix(filename: &str) -> Option<NaiveDate> {
     if filename.len() < 10 {
         return None;
     }
-    NaiveDate::parse_from_str(&filename[..10], "%Y-%m-%d").ok()
+    NaiveDate::parse_from_str(&filename[..filename.floor_char_boundary(10)], "%Y-%m-%d").ok()
 }
 
 fn is_older_than(path: &Path, cutoff: SystemTime) -> bool {