fix: zero address change

[shiva-istari]

Description
When a Zero node is bootstrapped, its --my address is permanently recorded in the Raft WAL via a ConfChangeAddNode entry. On every subsequent restart, WAL replay restores that original address into MembershipState, overwriting the current --my flag. This causes Alphas to connect to a stale address (e.g., localhost:5080 from an initial bulk load) even when Zero is restarted with a production FQDN.
This PR introduces a leader-driven reconciliation loop that detects mismatches between the live --my address and what is stored in MembershipState, and proposes a ConfChangeUpdateNode through Raft to durably correct the address across all nodes. An in-memory override in the Connect RPC ensures Alphas receive the correct address immediately, even before the Raft entry is committed.
fixes #9676

[matthewmcneely]

Smoke Tests on k8s

Verify the failure: bulk load -> k8s deploy reproduces the stale address bug

Ran a standalone Zero locally with --my=localhost:5080, bulk loaded the 21M movie dataset (3 shards), then injected the Zero WAL and Alpha p directories into a fresh Docker Desktop k8s cluster (1 Zero, 3 Alphas, sharded config, dgraph/dgraph:latest v25.3.3). When the k8s Zero started, WAL replay overwrote its live --my address with localhost:5080 — confirmed via /state. All three Alphas got stuck in a connection-refused loop trying to dial localhost:5080 inside their containers and never became Ready.

Verified fix works: bulk load -> k8s deploy with dgraph/dgraph:local (built on shiva/fqdn)

Rebuilt dgraph/dgraph:local from the PR branch (shiva/fqdn), redeployed the k8s cluster, and injected the same stale Zero WAL (localhost:5080) and bulk-loaded Alpha data from the bug repro. On startup, Zero's reconciliation loop detected the mismatch within seconds of leader election and corrected it via Raft:

Zero 0x1 address mismatch: MembershipState has "localhost:5080",
  expected "dgraph-dgraph-zero-0.dgraph-dgraph-zero-headless.dgraph.svc.cluster.local:5080".
  Proposing ConfChangeUpdateNode.
Applied ConfChangeUpdateNode for Zero 0x1:
  addr="dgraph-dgraph-zero-0.dgraph-dgraph-zero-headless.dgraph.svc.cluster.local:5080"

All three Alphas came up Ready, connected to Zero at the corrected address, and served queries against the 21M movie dataset across all 3 shard groups. This is the same stale WAL that caused a total cluster failure on main — the fix resolves it without manual intervention.