test(scheduler): raise interloper-scheduler coverage from 90% to 100%

[aaaaahaaaaa]

Closes the last 62 uncovered statements in interloper-scheduler. Every module is now at 100%. Tests only: no source file changes.

Coverage

Module	Before	After
executor.py	76%	100%
reaper.py	78%	100%
renewal.py	93%	100%
controller.py, hooks.py	95%	100%
cron.py, launcher.py	98%	100%
queue.py	100%	100%

The layout already mirrored the package one-to-one (8 flat modules, 8 test files), so there was nothing to move.

What the gaps actually were

Both real gaps — the executor and the reaper — were failure machinery, not happy paths: the code that decides whether a stuck run gets unstuck. That's now covered:

• a run that cannot be executed (missing, or with no component) is skipped rather than half-started
• a component whose kind declares no workload fails the run instead of silently succeeding, and the failure is reported as a RUN_FAILED event
• both halves of the reaper's failure record survive the other one raising — a broken event save still leaves the run marked failed, which is what unsticks it, and a broken complete is logged rather than escaping
• a launcher whose describe_run throws falls through to the dispatch timeout rather than leaving the run dispatched forever

The rest is the platform side of graph assembly: the transitive upstream walk that joins dependencies as non-materializable context (loaded once, never recomputed), and the failed-only retry that skips the lineage's prior successes while a whole-run retry recomputes everything.

Plus the reaper's hourly usage reconciliation (advisory only — and a failure there must not take the loop down), the controller's KeyboardInterrupt path, Launcher.describe_run's default of "cannot introspect", the hook sweep's first-tick watermark and its four _evaluate guards, and auto_renew defaulting on while skipping a connection whose config will not decode.

Two tests that needed care

A naive cron fire time. croniter normally returns aware datetimes; a naive one must be anchored in the job's zone, not read as UTC. The test asserts 02:00 Berlin → 00:00 UTC and carries a # noqa: DTZ001, because producing a naive datetime is precisely the branch under test.

A job with no cron expression. My first attempt gave it config={} and the branch stayed uncovered — the sweep filters on config["enabled"], so the job was never selected to be skipped. It needs {"enabled": True} with no cron.

Fake-store typing

_RecordingStore stands in for Store at 17 construction sites. Rather than 17 # ty: ignore comments, there's one _executor(store) helper that casts once — same spirit as the cast(...) approach in the api and db suites.

Verification

uv run ruff check, uv run ty check, uv run pytest packages/interloper-scheduler (111 passed, 100%) and the full workspace suite (2574 passed, 2 skipped) all green; the scheduler suite is stable across three runs. Pre-commit ran the same three hooks on the commit.

By Digitl

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!