health_checker: clear last_hc_http_status_ on network failures

[MyUmmaGumma]

Commit Message: health_checker: clear last_hc_http_status_ on network failures so the HDS EndpointHealthResponse does not ship a stale code from the last successful response after the upstream becomes unreachable.

Additional Description:

The HTTP health checker records the response status in host_->setLastHealthCheckHttpStatus() in HttpActiveHealthCheckSession::onResponseComplete(). The field is consulted by HdsDelegate::sendResponse() and attached to the HDS report's health_metadata as http_status_code. It is not cleared on subsequent probe failures, so once an upstream has had a single successful HTTP response, the cached status persists forever — even across network-level probe failures (connection refused, timeouts, goaway, reset stream).
An HDS server that interprets http_status_code for health states then sees:

t1: probe succeeds          → metadata.status_code = 200
t2: probe fails at network  → metadata.status_code = 200 (stale)

This is asymmetric with the HealthStatus enum, which reflects the latest probe outcome. The fix is to clear last_hc_http_status_ in handleFailure() when the failure type is NETWORK or NETWORK_TIMEOUT, restoring the contract that http_status_code in health_metadata represents the response code from the most recent HTTP exchange (if any), absent otherwise.

Does not affect ACTIVE failures (where the upstream returned a non-2XX response code): in that case setLastHealthCheckHttpStatus() has already been called in HttpActiveHealthCheckSession::onResponseComplete() with the actual response code, which is current and correct.

Risk Level: Low

Narrowly-scoped change in the base health-checker session's failure handler. No API changes. New unit test exercises the cleared-on-timeout case.

Testing: Unit

Added HttpHealthCheckerImplTest.LastHealthCheckHttpStatusClearedOnNetworkFailure that verifies the recorded HTTP status is cleared after a timeout-driven probe failure following a successful 200 response.

Docs Changes: No

Release Notes: Added entry under bug_fixes: in changelogs/current.yaml:

- area: upstream
  change: |
    health_checker: clear cached HTTP response status code on network-level
    health check failures so the HDS report does not ship a stale code after
    the upstream becomes unreachable.

Platform Specific Features: N/A

[repokitteh-read-only]

Hi @MyUmmaGumma, welcome and thank you for your contribution.

We will try to review your Pull Request as quickly as possible.

In the meantime, please take a look at the contribution guidelines if you have not done so already.

🐱

Caused by: #45687 was opened by MyUmmaGumma.

see: more, trace.

[mathetake]

This is a follow up on #43804

[mathetake]

/retest