fix: standardize structured log event names to kebab-case

[rcbevans]

Summary

The codebase used 4+ different naming conventions for structured log event names. The dominant convention (~40+ events) is kebab-case (conn-close-error, pool-draining, notify-payload-parse-failed). This PR standardizes the 15 non-conforming events to kebab-case.

Breaking change

This is a breaking change for log pipelines that filter on event names. Any dashboards, alerts, or log routing rules that match on the old event names will need to be updated.

Renamed events

Old name	New name	File
"prune completed"	"prune-completed"	_leader_sweeps.py
"prune failed"	"prune-failed"	_leader_sweeps.py
"archive expiry completed"	"archive-expiry-completed"	_leader_sweeps.py
"applied migrations before startup"	"migrations-applied-before-startup"	migrate.py
"no pending migrations"	"no-pending-migrations"	migrate.py
"workgroup.child_exit"	"workgroup-child-exit"	workgroup.py
"workgroup.shutdown_signal"	"workgroup-shutdown-signal"	workgroup.py
"workgroup.health_kill"	"workgroup-health-kill"	workgroup.py
"workgroup.background_task_failed"	"workgroup-background-task-failed"	workgroup.py
"workgroup.shutdown_begin"	"workgroup-shutdown-begin"	workgroup.py
"workgroup.shutdown_complete"	"workgroup-shutdown-complete"	workgroup.py
"workgroup.burst_limit_exceeded"	"workgroup-burst-limit-exceeded"	workgroup.py
"job_failed"	"job-failed"	_handlers.py
"watch_reclaims-visibility-risk-probe-failed"	"watch-reclaims-visibility-risk-probe-failed"	_taskq.py
"watch_reclaims-listen-connection-lost"	"watch-reclaims-listen-connection-lost"	_taskq.py

What was changed

• Source files (5): Event name string literals in logger.info(), logger.warning(), logger.error(), and logger.critical() calls. Also updated inline comments and docstrings that referenced the old event names.
• Test files (3): Assertions and test docstrings that check for the renamed event names.
• No variable names, function names, or code logic was modified.

What was NOT changed

• typer.echo("no pending migrations") in cli.py — this is a CLI output string, not a structured log event name.
• watch_reclaims-visibility-delay-at-risk and watch_reclaims-listen-reconnected — these events already use the watch_reclaims snake_case prefix but were not in scope for this audit. They can be addressed in a follow-up if desired.

Verification

• uv run ruff check . — all checks passed
• uv run ruff format --check . — 475 files already formatted
• uv run pyright (on changed source files) — 0 errors, 0 warnings
• uv run pytest tests/test_notify.py tests/test_leader.py tests/test_leader_prune.py tests/test_shutdown_orchestrator.py tests/test_migrate_coverage.py tests/test_migrate_runner.py tests/test_workgroup.py tests/test_consumer.py tests/test_watch_reclaims.py -v --tb=short -m "not integration" — 267 passed, 35 deselected
• Final grep for all 15 old event names across src/, tests/, docs/ — zero hits (only the intentionally-preserved typer.echo CLI output remains)

[rcbevans]

CI failure is the pre-existing flaky test test_pg_drop_during_phase2 (asserts cancelled but gets abandoned under timing pressure). Same flake noted in PR #39. Retriggered.