ailoop v0.2.1

Fixes

ailoop resume now establishes an explicit run intent for paused and stopped loops.
Loop state and control transitions are transactional, so newer pause or stop requests win
before a detached dashboard worker can claim the next iteration.
State writes use durable atomic replacement, while event appends are flushed and synced.

On POSIX, catchable runner exits clean and reap the original process group before the loop
becomes resumable.
Unconfirmed cleanup records cleanup_failed and blocks rerun, resume, control, and removal;
it does not claim that the process was terminated.
Abrupt parent SIGKILL can leave a runner alive and require manual cleanup. The persisted loop
remains active and fails closed instead of authorizing duplicate work.
Removed loop IDs are durably retired and cannot be reused by delayed workers or events.
Scheduled loops remain configuration-only until scheduler execution is implemented.

Added lifecycle-safety coverage for control races, durable claims, cleanup failures,
process-group teardown, retired IDs, removal races, and scheduled-loop guardrails.
CI passes on Python 3.11 and 3.13.
272 tests pass with 83% coverage.

Download the wheel attached to this GitHub release, then run:

pipx install ./ai_loop-0.2.1-py3-none-any.whl
ailoop --help

PyPI publication is not part of this release.