v0.11.0

What's Changed

• docs(changelog): 0.11.0 — model & voice discovery, reasoning, SDK updates
• style(test): ruff-format test_openai.py
• fix(azure): add the sampling config fields Azure inherits from OpenAI
• chore(example): default the local realtime example to gpt-realtime-2
• refactor(openai): dedupe sampling kwargs; quiet dynamic-kwargs overload
• fix(anthropic): capture and round-trip thinking block signatures
• fix(openai): drop reasoning_effort when the turn carries tools
• fix(anthropic): adaptive thinking + drop temperature for newer models
• feat(openai): map reasoning_effort for OpenAI reasoning models
• fix(openai): omit temperature for models that only accept the default
• fix(openai): send max_completion_tokens for OpenAI's newer models
• fix(gemini): record usage and round-trip thought_signatures with tools
• fix(mistral): emit canonical input_tokens/output_tokens usage keys
• feat(ai): ModelInfo.capabilities + live Ollama capability probe
• style(voice): drop "legacy" from voice-catalog docstrings
• feat(voice): realtime voice catalog (available_voices/list_voices)
• fix(neonize): guard the event_global_loop workaround for neonize 0.3.18
• feat(ai): surface reasoning/thinking across all providers
• test: fix order-dependent event-loop tests; restore mistral import guard
• chore(deps): update AI/provider SDKs to latest
• chore(deps): update mistralai SDK to 2.x
• feat(ai): provider model discovery + image data-URI handling

Full Changelog: v0.10.0...v0.11.0