checkpoint: pre list-to-vector refactor, fixes dictionary crash, mostly

- Add uncompressed dictionary (.dict) file support to avoid decompression memory issues
- Implement chunked on-demand parsing for large definitions
- Add backward navigation with re-parse capability
- Limit cached pages to MAX_CACHED_PAGES (4) to prevent memory exhaustion
- Add helper script for extracting/recompressing dictzip files

lib/StarDict/StarDict.h

@@ -6,7 +6,7 @@
 #include <string>
 
 // StarDict dictionary lookup library
-// Supports .ifo/.idx/.dict.dz format with linear scan lookup
+// Supports .ifo/.idx/.dict (uncompressed) and .ifo/.idx/.dict.dz (compressed) formats
 class StarDict {
  public:
   struct DictInfo {
@@ -38,16 +38,22 @@ class StarDict {
   };
   DictzipInfo dzInfo;
 
+  // Whether to use uncompressed .dict file (preferred) or compressed .dict.dz
+  bool useUncompressed = false;
+
   // Parse .ifo file
   bool loadInfo();
 
-  // Load dictzip header for random access
+  // Load dictzip header for random access (only if using compressed)
   bool loadDictzipHeader();
 
   // Read word at given index file position, returns word and advances position
   bool readWordAtPosition(FsFile& idxFile, uint32_t& position, std::string& word, uint32_t& dictOffset,
                           uint32_t& dictSize);
 
+  // Read definition directly from uncompressed .dict file (no decompression needed)
+  bool readDefinitionDirect(uint32_t offset, uint32_t size, std::string& definition);
+
   // Decompress a portion of the .dict.dz file
   bool decompressDefinition(uint32_t offset, uint32_t size, std::string& definition);