1.0.2 — Fix settings loading

Fix settings loading from TokensCounter.sublime-settings, so package defaults and user overrides for model_name and tokenizer_encoding are applied correctly. Resolves #2.

The configured default remains gpt-4o (using o200k_base). To select an encoding explicitly, set "model_name": null; a non-empty model name takes precedence. Verified in Sublime Text 4214 with real tiktoken: defaults, both encodings, model precedence, and counter display.