feat(failover): elect the highest-offset replica as master

[binhnguyenduc]

What

Changes the fallback master election (used when there is no clear existing master — bootstrap or all-down recovery) to promote the replica with the highest replication offset instead of the oldest pod.

Why

The oldest pod is an arbitrary choice: it may be a replica that has fallen behind, so promoting it silently discards the writes held by more up-to-date replicas. Electing by master_repl_offset promotes the instance with the most data, minimizing data loss on failover.

Behaviour

• Query master_repl_offset (via INFO replication) for each pod once, then order by offset desc, creationTimestamp asc.
• The oldest pod remains the tie-breaker (equal offsets) and the fallback when offsets can't be read — so on a fresh/idle cluster the result is identical to before.
• Unreachable pods sort last (offset treated as -1) and are never elected (they'd fail promotion anyway).

Design note

The interface method keeps its name SetOldestAsMaster (oldest is now the tie-breaker/fallback) to bound the blast radius on this failover-critical path — the doc comment describes the offset-first behaviour.

Changes

• service/redis/client.go: GetReplicationOffset + regex + metric; interface + mock.
• operator/redisfailover/service/heal.go: offset-based ordering in the election.
• Tests: existing election tests get equal-offset mocks (oldest tie-breaker preserved); new TestSetOldestAsMasterPrefersHighestOffset proves highest-offset wins.

Upstream references

• Make master by the offset  spotahome/redis-operator#385 (make master by the offset)

Test

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