v0.7.0

Add stateful sessions, model-native thinking, prompt-prefix reuse, and broader JSON boundary handling.

SwiftPM binary artifact for CarbocationLocalLLM's llama runtime.

This archive redistributes static llama.cpp/ggml object code through
libllama-combined.a. See THIRD_PARTY_NOTICES.md in this release tag
for bundled and linked notices.

Checksum:
6da36f751e1398acd7fc010400332c2e92452db366f436a99d481ec8e22dc9e3