Scheduler logs SUCCESS when a job returns ok:false (soft failure)

[jung-thomas]

Bug

A scheduled job whose runner returns { ok: false, error } (rather than throwing) was recorded as SUCCESS in PipelineLog and JobLastRun. The scheduler only branched on a thrown exception; a soft-failure return value fell through to the success path.

Concretely for semaphore-tag-sync: a fail-shut fetch/mapping error returns { ok:false, error } by design (so nothing is written), yet the run showed green — masking that the taxonomy import silently did nothing. This is why the "no new tags imported" prod symptom was invisible in the run history.

Fix (in this PR)

srv/jobs/scheduler.js now inspects the runner's return value after fn(logId):

• result && typeof result === 'object' && result.ok === false → outcome error, logPipelineEnd(logId, 'FAILED', summary, errorMessage), and a ScheduledJobFailed alert is raised.
• otherwise → logPipelineEnd(logId, 'SUCCESS', summary) as before.

errorMessage derives from result.error when present, else <jobName> returned ok:false.

Acceptance

• A runner returning {ok:false} produces a FAILED PipelineLog row with the error in the error column (arg 4 of logPipelineEnd, not the summary column) and a failed JobLastRun.
• A thrown exception still fails as before.
• A runner returning {ok:true} (or a plain summary) still succeeds.
• Covered by the scheduler unit suite.

Ref: #2184.