Description

All sources currently refresh simultaneously at t=0 on every startup and every 30s interval, causing a thundering herd against the Next.js API. Add deterministic per-source jitter so each source's first fetch is offset by hash(source_kind) % MAX_JITTER_MS milliseconds. The offset should be stable across restarts (same source = same offset), so phase relationships don't change between runs.

Implementation sketch:

• Write a _jitter(kind: str, max_ms: int = 8000) -> float function using hashlib.md5(kind.encode()).digest() as seed
• In on_mount, instead of self.set_interval(REFRESH_SECS, self.refresh_data), fire each source's first refresh after its individual jitter offset via self.set_timer(jitter_secs, lambda: ...) then start the repeating interval
• Or: a single interval timer that checks per-source "next due" timestamps rather than refreshing everything at once

Acceptance Criteria

• #1 Restarting the TUI does not cause all API calls to fire at the same instant
• #2 Same source always gets the same jitter offset across restarts
• #3 Max jitter is configurable via a constant at the top of the file
• #4 Overall data refresh still completes within REFRESH_SECS of startup