- Update `line1` and `line2` assignments to begin at 1 instead of 0
- Add defensive checks for line numbers exceeding file content length
- Modify debug logging to reflect updated line numbers
- Refactor tests and examples for consistent use of 1-based indexing

Convert HashMap to AHashMap using .into_iter().collect() to match
the expected type for tokenizers::WordPiece::builder().vocab().
This fixes compilation error E0277.

Use .into() to let the compiler convert the array into the appropriate
type (HashMap for local tokenizers 0.21.1, AHashMap for CI tokenizers 0.21.4).
This ensures compatibility across different tokenizers versions.

- Update tokenizers from 0.21.1 to 0.21.4 to match CI version
- Simplify vocab initialization using array literal without .into()
- This ensures consistent behavior across all build environments

- Normalize CRLF to LF in all parser tests before parsing
- This ensures consistent byte offsets across Windows and Unix platforms
- Fixes test failures in test_kotlin_main and test_kotlin_person on Windows
- The issue was that reference JSON files use LF-based byte offsets