v0.1.0 — first release

First published release. Extracted from a real, months-long production use orchestrating
DeepSeek translation waves (10 concurrent workers, hundreds of documents, real dollar caps that
were never exceeded) — not written speculatively ahead of real experience.

What's in it

• wave.py — budget-aware concurrent wave scheduler: a concurrency cap AND a cumulative
  dollar-budget cap, real token/cost tracked per job via run_pi.py's ledger integration
  (embedded, not a separate step a caller has to remember).
• run_pi.py — start_pi/collect_pi_result/run_pi for spawning and tracking pi
  processes, with optional systemd-run --user --scope memory isolation (CgroupLimits) so one
  runaway job's OOM can't take the whole terminal session down with it (a real incident this
  fixed). resume_pi_headless() for continuing a closed session's context in a new process.
• pi-wave-job (CLI) — one ad-hoc job from the command line, now the default way to launch
  any Pi/agent job across every project using this library: uv tool install --editable . puts
  it on PATH. Structurally immune to a real bug class (pi_job.sh's file-staging race) that
  affected the shell-script predecessor it replaces.
• interactive.py / pi-wave-session (CLI) — launch a long-running, remotely-steerable
  interactive session (tmux/screen + pi-messenger mesh), separate from the one-shot job case.
• escalate.py — escalation protocol for a wave to flag something needing human attention.
• stage_check.py — generic stage-check framework for multi-step pipelines.
• coordinator.py — the "who's watching the wave" role, confirmed working with either an
  interactive Pi session or a non-Pi agent (Claude Code or otherwise) reading the same
  escalation/summary files directly.
• plugins/stt_translation — the real production pipeline this library was extracted from,
  bundled as reference/example data.

108 tests passing. See README.md for usage, CONTRIBUTING.md for the commit-message
convention this repo enforces via a pre-commit hook.