feat: cap job_runs history at last 10 per job

Tailscale, Patchmon, and Semaphore sync jobs all wrote into a shared
job_runs table with no retention. With default poll intervals of 15-60
minutes, history grew unbounded.

- Add pruneJobRuns(jobId) and pruneAllJobRuns() helpers.
- Prune after every completeJobRun() so new runs trim old ones.
- Prune once on init() to clean up existing over-cap rows.
- Prune in importJobs() so re-imported runs are also capped.
- Defensive LIMIT 10 in getJobRuns() for the read path.

No UI changes needed — _renderRunList already renders whatever the
server returns. No schema migration — only row deletions.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

tests/db.test.js

@@ -450,4 +450,17 @@ describe('job_runs', () => {
     expect(runs[0].id).toBe(r2);
     expect(runs[1].id).toBe(r1);
   });
+
+  it('caps history at the last 10 runs per job', () => {
+    createJob(baseJob);
+    const id = getJobs()[0].id;
+    for (let i = 0; i < 15; i++) {
+      const runId = createJobRun(id);
+      completeJobRun(runId, 'success', `run ${i}`);
+    }
+    const runs = getJobRuns(id);
+    expect(runs).toHaveLength(10);
+    expect(runs[0].result).toBe('run 14');
+    expect(runs[9].result).toBe('run 5');
+  });
 });