helper: Bound internet checks so a DNS hang is not INTERNET_CHECK_FAIL

[joaoantoniocardoso]

Summary

Helper's internet check waited for every website probe, including DNS lookups that ignore the 5s socket timeout. A lost UDP query to 8.8.8.8 costs glibc ~10s, which matches the frontend's 10s Axios deadline, so the UI showed INTERNET_CHECK_FAIL while ICMP still worked.

• check_internet_access now returns within 4s. Sites still resolving are online: false, error: "timeout" — not a failed network. In-flight probes are reused so a hung DNS lookup cannot queue a second worker.
• Website probes use HEAD. Per-site results cache for 20s; the overall check caches for 25s to cover overlapping UI polls (20s).
• Service scans skip Zenoh REST/scout (7117/7447). In-flight temporary_cache calls coalesce per args.
• The unused periodic internet check is gone. Factory-mode HTTP still runs off the event loop.
• The UI keeps the last verdict until three polls in a row fail, and does not treat Helper budget timeouts as offline. web_services polling is 10s so scans no longer stack every 5s.

Fixes #4176

Test plan

• Open BlueOS with working internet: globe is online, notifications stay empty
• GET /helper/latest/check_internet_access returns in well under 10s (Pi retest: 200 in ~19ms, all five sites online; cached/concurrent polls 12–25ms)
• Helper logs no 1s timeouts on Zenoh 7447/7117 after restart
• A single timed-out Axios poll does not clear has_internet or raise INTERNET_CHECK_FAIL
• All-timeout Helper responses do not flip the UI to OFFLINE