fix: mongodb leader fallback and single-member reconfig

[wallyxjh]

Summary
Fix MongoDB scale-in failures (cluster has no leader) in KB 0.9.x by adding leader fallback, single-member reconfig, and join-member de-dup logic in lorry.

What changed

• Fallback to replset client when DCS leader lease is missing/expired.
• Force reconfig after scale-in to a single member to ensure electable primary.
• De-duplicate hosts on join to avoid NewReplicaSetConfigurationIncompatible.

Why
In KB 0.9.x the MongoDB HA loop is disabled, so leader leases are not renewed. Scale-in operations that depend on a leader fail and can leave the cluster stuck in Updating. Single-member replicasets can also
remain SECONDARY without a forced reconfig.

Scope

• Affects MongoDB lorry logic only.
• No changes to MySQL/PG/Redis.

Validation

• Scale-in from 3→1 completes without cluster has no leader.
• Single-member replset becomes PRIMARY.
• Scale-out/join no longer fails with duplicate host errors.