fix: prevent false service error alerts during operations

[Pyatakov]

Description

• Fixed a false "Something went wrong" modal that appeared during long operations such as policy import or dry-run, even though the operation completed successfully.
• Root cause was in the api-gateway: it checked each service's health by asking for a status reply and ignoring anything that arrived later than 300ms. A service that is merely busy answers late, so it looked exactly like a crashed one and was reported as unavailable, triggering the modal mid-operation.
• The gateway now tracks the most recent status reply from each service, including late ones, and reports a service as unavailable only after it has been silent for longer than two heartbeats (~70s). A busy service keeps answering (late), so it is no longer mistaken for a down service, while a genuinely down service is still detected.
• The frontend re-checks status across a short grace period before showing the modal, so a brief blip (such as a service restarting) is ignored.
• The alert resets once services recover, so a later genuine outage is reported again (previously it only fired once).
• Pending checks are cancelled when the websocket connection closes, so an old timer cannot trigger the modal during a reconnect.

Resolves #6281

[github-actions]

Test Results

 32 files  ±0   64 suites  ±0   6m 17s ⏱️ +47s
 35 tests ±0   32 ✅  - 3  0 💤 ±0  3 ❌ +3 
165 runs  ±0  162 ✅  - 3  0 💤 ±0  3 ❌ +3 

For more details on these failures, see this check.

Results for commit fb7467c. ± Comparison against base commit 89b8867.

♻️ This comment has been updated with latest results.