rgw/rest: track connection failures per-IP instead of per-endpoint

rgw/rest: track connection failures per-IP instead of per-endpoint

Previously, when a connection to a zone endpoint failed, the entire
endpoint was marked as unavailable for a timeout period. Since we now
resolve endpoints to all their IP addresses (via DNS A/AAAA records),
we can be more granular: track failures at the individual IP level.

Introduce ResolvedIP struct that pairs each IP's connect_to string
with its own failure timestamp. When selecting an IP for a request,
round-robin skips IPs that have recently failed, allowing traffic to
continue flowing to healthy nodes even when some are down.

An endpoint-level last_failure_time is maintained as a fast-path
optimization to avoid scanning all IPs when none have failed recently.

Signed-off-by: Oguzhan Ozmen <oozmen@bloomberg.net>