v0.1.0

Initial release. Runs one review across multiple local LLM CLI tools in
parallel and synthesizes their outputs, with three human checkpoints.

Added

• review <plan.md> command: assemble prompt from rubric + plan, fan out
  reviewers in parallel, synthesize, and persist the run.
• Config-driven reviewers: adding a reviewer is a config.json edit, one
  generic CLI-reviewer path (no per-provider code).
• Parallel execution via async Process::start() + polling, capped at
  max_parallel; a single failing reviewer never aborts the run.
• Synthesizer that tags findings as consensus / contested / singleton via
  config/synthesis.md; optional arbiter step (may be disabled).
• Three interactive checkpoints (confirm prompt, review raw outputs, accept
  synthesis); --yes skips them.
• --dry-run prints the resolved reviewer commands without spawning anything.
• --config to point at an alternate config file; paths resolve relative to
  the config file's directory.
• "Re-run all failed" option at checkpoint 2.
• Configurable process poll interval.
• Read-only reviews: each agentic CLI is passed its read-only switch.
• Run persistence to <output_dir>/<timestamp>/.
• Tolerant JSON parser for reviewer and synthesis output; malformed output is
  stored as unstructured and surfaced at checkpoint 2.
• Astro Starlight documentation site under /website.