fix(failover): heal on a running quorum instead of requiring all pods

[binhnguyenduc]

What

Relaxes the CheckAndHeal gate: the operator now proceeds to heal when a quorum (majority) of redis and sentinel pods are Running, instead of requiring the full set.

Why

CheckAndHeal returned early unless every redis and sentinel pod was Running. A single pod stuck Pending — unschedulable anti-affinity, an availability-zone loss taking a node down — blocked all healing: no master election, no sentinel monitor reconfiguration. The cluster could not recover on its own.

• Issue with #533 spotahome/redis-operator#535 / Unsatisfiable Affinity rule causes the entire redis cluster to be down, and redis-operator fails to recover even when the CR is reverted to correct Affinity rule spotahome/redis-operator#552: an unsatisfiable affinity rule leaves a pod Pending and bricks recovery.
• Sentinel misconfigured after availability zone failure spotahome/redis-operator#601: after an AZ failure a freshly-scheduled sentinel keeps its 127.0.0.1 monitor and is never reconfigured to the new master.

Why it's safe

The downstream heal path already operates only on running/reachable pods (GetRedisesIPs filters to PodRunning; SetMasterOnAll/CheckAllSlavesFromMaster skip unreachable pods), and the operator's force-election stays guarded by CheckSentinelQuorum + CheckIfMasterLocalhost. So only the early-return gate needed to change — no downstream behaviour changes.

• Quorum = replicas/2 + 1 (2-of-3 proceeds; 1-of-3 still waits, so a mostly-down cluster is not auto-healed).
• The normal all-running case behaves identically.
• Bootstrap mode still waits for the full set (unchanged).

Changes

• check.go: AreQuorumRunning helper + IsRedisRunningQuorum/IsSentinelRunningQuorum predicates (interface + mock).
• checker.go: the CheckAndHeal gate uses the quorum predicates.
• Tests: AreQuorumRunning table test; existing CheckAndHeal gate test routed to the quorum predicates.

Upstream references

• Issue with #533 spotahome/redis-operator#535, Unsatisfiable Affinity rule causes the entire redis cluster to be down, and redis-operator fails to recover even when the CR is reverted to correct Affinity rule spotahome/redis-operator#552 (unschedulable pod blocks recovery)
• Sentinel misconfigured after availability zone failure spotahome/redis-operator#601 (sentinel misconfigured after AZ failure)

Test

go build ./..., go test ./..., gofmt clean.