v0.2.1

✨ What's New

• Added markdown artifact normalization and non-speakable symbol stripping to the preprocessing pipeline to improve speech input robustness for mixed-format text.
• Introduced token-aware chunking controls (target_min_tokens, target_max_tokens, absolute_max_tokens) across config, API models, and runtime option resolution for more predictable chunk sizing.
• Improved server chunk processing by deferring unspeakable validation to chunk cleanup when splitting and adding safer fallback/recovery behavior during synthesis.
• Expanded and aligned test coverage for preprocessing/chunking behavior, and verified the full suite including integration tests (45 passed).

Full Changelog: v0.2.0...v0.2.1