security: harden command exec, IPC, deep-link, cookies, and persistence

Seven-tier security audit of the main process, each finding fixed with a
regression test. Typecheck (node + web) clean; unit tests 106 -> 140.

- Tier 1 (command exec/argv): allowlist the yt-dlp --update-to channel
  (blocks arbitrary-binary-install RCE); gate per-download extraArgs behind
  the customCommandEnabled consent flag in main (blocks --exec RCE); resolve
  taskkill/schtasks by absolute System32 path; validate per-download
  outputDir; normalize the URL in assertHttpUrl and use it at every spawn.
- Tier 2 (input validation): fix isSafeFilenameTemplate drive-relative
  ('C:foo') and Windows dotted-'..' traversal bypasses; percent-encode the
  untrusted id in entryUrl; catch reserved device names with extensions in
  sanitizeDirSegment.
- Tier 3 (fs/backup): drop malformed template rows in importBackup;
  normalize deep-link URLs via assertHttpUrl.
- Tier 4 (cookies): confine the sign-in window's navigations/popups to web
  URLs (recursively) and deny all web permissions on its session.
- Tier 5 (deep-link/argv): bound the .url file read to 64 KB; match the
  aerofetch:// scheme case-insensitively.
- Tier 6 (Electron window): deny camera/mic/geolocation/USB/HID/serial/
  Bluetooth permissions on the app window.
- Tier 7 (network/persistence): restrict the watched-source RSS fetch to
  youtube.com feed URLs (SSRF guard); complete isValidSource validation.

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

test/deeplink.test.ts

@@ -42,6 +42,20 @@ describe('extractIncomingUrl — aerofetch:// protocol', () => {
   it('returns null when there is no url= param', () => {
     expect(extractIncomingUrl(['AeroFetch.exe', 'aerofetch://download'])).toBeNull()
   })
+
+  it('normalises the target, stripping embedded control chars (audit T3 / F5)', () => {
+    // A tab spliced into the inner URL must not survive to the renderer banner.
+    const raw = 'https://www.youtube.com/watch?v=abc\tdef'
+    const arg = `aerofetch://download?url=${encodeURIComponent(raw)}`
+    expect(extractIncomingUrl(['AeroFetch.exe', arg])).toBe(
+      'https://www.youtube.com/watch?v=abcdef'
+    )
+  })
+
+  it('matches the aerofetch:// scheme case-insensitively (audit T5)', () => {
+    const arg = `AEROFETCH://download?url=${encodeURIComponent(TARGET)}`
+    expect(extractIncomingUrl(['AeroFetch.exe', arg])).toBe(TARGET)
+  })
 })
 
 describe('extractIncomingUrl — .url Internet Shortcut (Explorer "Send to")', () => {
@@ -62,6 +76,18 @@ describe('extractIncomingUrl — .url Internet Shortcut (Explorer "Send to")', (
   it('ignores files that merely end in .url-like text but are not real paths', () => {
     expect(extractIncomingUrl(['AeroFetch.exe', 'not-a-real-path.url'])).toBeNull()
   })
+
+  it('reads a URL= line that falls within the 64 KB size cap (audit T5)', () => {
+    const junk = '; padding\r\n'.repeat(100) // ~1 KB of leading content
+    const path = urlFile('Small.url', `[InternetShortcut]\r\n${junk}URL=${TARGET}\r\n`)
+    expect(extractIncomingUrl(['AeroFetch.exe', path])).toBe(TARGET)
+  })
+
+  it('ignores a URL= line that falls past the 64 KB size cap (audit T5)', () => {
+    const junk = '; padding\r\n'.repeat(8000) // ~88 KB, beyond the read window
+    const path = urlFile('Huge.url', `[InternetShortcut]\r\n${junk}URL=${TARGET}\r\n`)
+    expect(extractIncomingUrl(['AeroFetch.exe', path])).toBeNull()
+  })
 })
 
 describe('extractIncomingUrl — no match', () => {