Release v0.1.82

chore(release): prepare v0.1.72

Release v0.1.71

Merge pull request #19 from lerim-dev/feat/autoresearch-optimization

feat: v0.1.70 — AutoResearch optimization + eval infrastructure

Release v0.1.68

chore: bump version to 0.1.67

feat: add parallelism support for extraction windows, async explore, …

…and adaptive prompts

Three config knobs control concurrent execution:
- max_workers (extract): parallel window processing via ThreadPoolExecutor
- max_explorers (explorer): concurrent async explorer subagent calls
- parallel_pipelines (server): extract + summarize in same tool turn

Explorer tool changed from sync run_sync() to async await agent.run()
for true concurrent dispatch via PydanticAI asyncio.create_task.

Prompts adapt automatically: parallel text when knobs > 1, sequential
when = 1. Set all to 1/false for local models (Ollama/MLX).

Removed unused max_workers from [roles.summarize] — refine/fold is
inherently sequential.