fix: preserve sentence punctuation in chunk_text, align clean_text defaults, fix version

[voidborne-d]

Summary

Three interrelated fixes for text processing and API consistency.

1. chunk_text destroys sentence-ending punctuation (relates to #67, #72)

Root cause: re.split(r'[.!?]+', text) splits on sentence-ending punctuation and removes it entirely. The resulting chunks lose all periods, question marks, and exclamation marks. ensure_punctuation() then adds a comma to each chunk, so the TTS model sees commas where periods/question marks should be — degrading prosody, intonation, and causing the model to treat declarative, interrogative, and exclamatory sentences identically.

This contributes to:

• BUG:[ONNXRuntimeError] : 2 : INVALID_ARGUMENT #67 — degraded output quality for longer texts (punctuation-dependent prosody is lost across all chunks)
• 合成语音末尾容易吞音 #72 — end-of-sentence audio issues (model receives comma instead of period/question mark)

Fix: Replace the destructive re.split(r'[.!?]+', text) with re.split(r'(?<=[.!?])\s+', text) — a lookbehind pattern that splits on whitespace after sentence-ending punctuation, keeping the punctuation attached to the preceding sentence.

2. clean_text default mismatch between public API and internal model

Users calling the public API (KittenTTS) never get text preprocessing (number expansion, abbreviation handling, etc.) unless they explicitly pass clean_text=True, because KittenTTS.generate() defaults to clean_text=False while the internal KittenTTS_1_Onnx.generate() defaults to True. Additionally, generate_to_file() didn't expose clean_text at all.

Fix: Align all KittenTTS wrapper defaults to True, add clean_text parameter to generate_to_file(), remove stale debug print() from generate().

3. Version mismatch (#80)

__init__.__version__ says 0.1.0 while pyproject.toml and setup.py say 0.8.1, causing wheel installation failures.

Fix: Update __init__.__version__ to 0.8.1.

Tests

26 new regression tests (tests/test_chunk_text_and_api.py), all lightweight (no model/ONNX/GPU/espeak needed):

• TestChunkTextPunctuationPreservation (5)
• TestChunkTextLongSentences (2)
• TestChunkTextEdgeCases (6)
• TestEnsurePunctuation (6)
• TestCleanTextDefaults (3)
• TestVersionConsistency (2)
• TestSourceAudit (2)

26 passed, 0 failed