Self-review: drop the formatter re-export shim, import @shared/format directly

The prior commit kept a pass-through `export { fmtBytes, fmtSpeed, fmtEta }` in
main/lib/formatters.ts purely so existing importers wouldn't have to change —
which re-exported fmtSpeed (zero importers, dead) and left download.ts re-
exporting fmtBytes/fmtEta with no consumer at all, undercutting the "one home"
goal.

Now the real consumers import from the canonical module directly:
- probe.ts and the download.test import from @shared/format.
- formatters.ts keeps only its internal `import { fmtBytes }` (for parseProgress)
  and no longer re-exports.
- download.ts re-exports only parseProgress (its own function); the dead
  fmtBytes/fmtEta re-export is removed.

No behavior change. typecheck + 248 tests + eslint + prettier green.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

src/main/download.ts

@@ -80,9 +80,9 @@ export function hasActiveDownloads(): boolean {
 }
 
 // --- Formatting helpers (raw yt-dlp numbers → human strings) ----------------
-// Implementations live in lib/formatters.ts (no electron import chain) so they
-// can be unit-tested in isolation (L37). Re-exported here for external callers.
-export { fmtBytes, fmtEta, parseProgress } from './lib/formatters'
+// parseProgress lives in lib/formatters.ts (no electron import chain) so it can
+// be unit-tested in isolation (L37). Byte/ETA formatters live in @shared/format.
+export { parseProgress } from './lib/formatters'
 import { parseProgress } from './lib/formatters'
 
 function send(wc: WebContents, ev: DownloadEvent): void {