v0.1.0

Added
Published on PyPI as llm-parry; imported as parry.
parry.install() hooks OpenAI chat completions, the OpenAI Responses API, Anthropic messages and Anthropic beta messages, on sync and async clients, including .with_raw_response and stream=True.
Parry guard: response cache (blake3 key, bounded LRU with TTL), per-model token-bucket rate limiting with tier-1 defaults, cost tracking with a dated pricing table, a max_cost_usd budget, and retry with exponential backoff, jitter and Retry-After.
Scoping: with Parry():, parry.wrap(client) and parry.bypass().
Optional SQLite persistence for the cache and the spend ledger, plus the parry stats / parry pricing CLI.
LangChain ParryCallback and LlamaIndex ParryHandler.
Standalone Rust primitives: Cache, RateLimit, Cost, Retry, count_tokens and estimate_tokens.