fix: re-gossip dead/suspect accusations on stale alive and unknown node

[emam07]

Problem

During mass cluster restarts, nodes can get permanently stuck as dead
in peers' views. The refutation mechanism never triggers because the
accusation (dead/suspect message) never reaches the restarted node.

Root cause — two silent drops in state.go:

Bug 1 (aliveNode, line ~1076): When a node receives
alive(node2, inc=1) but already holds dead(node2, inc=100), it
returns silently. It does not re-gossip the dead accusation back toward
the restarted node, so the node never calls refute().

Bug 2 (deadNode / suspectNode): When a dead or suspect message
arrives at a node that has never heard of the target (common in
freshly-joined nodes during a mass restart), it is silently dropped
instead of forwarded. This creates a gossip black hole that prevents
the accusation from propagating through nodes with incomplete views.

Both bugs together mean the restarted node broadcasts alive(inc=1)
indefinitely but no node ever sends back the dead(inc=100) accusation
it needs to refute. The node stays dead in affected peers' views
permanently — until an accidental TCP push/pull sync fixes it.

Reported in #311.

Fix

• aliveNode(): when a stale alive message is received for a dead or
  suspect node, re-queue the accusation for gossip so the restarted
  node can receive it and refute.
• deadNode() / suspectNode(): forward dead/suspect messages for
  unknown nodes rather than dropping them. The TransmitLimitedQueue
  already bounds retransmissions to RetransmitMult × log(N+1).
• Added [INFO] log lines when nodes are marked suspect or dead for
  operator visibility.

Tests

Four new tests in state_test.go:

Test	What it proves
TestMemberList_AliveNode_ReGossipsDeadAccusation	Stale alive re-queues dead accusation
TestMemberList_AliveNode_ReGossipsSuspectAccusation	Stale alive re-queues suspect accusation
TestMemberList_DeadNode_UnknownNode_ForwardsMessage	Dead msg forwarded for unknown node
TestMemberList_SuspectNode_UnknownNode_ForwardsMessage	Suspect msg forwarded for unknown node

All existing tests pass.

[hashicorp-cla-app]

All committers have signed the CLA.

[hashicorp-cla-app]

Thank you for your submission! We require that all contributors sign our Contributor License Agreement ("CLA") before we can accept the contribution. Read and sign the agreement

Learn more about why HashiCorp requires a CLA and what the CLA includes

_{Have you signed the CLA already but the status is still pending? Recheck it.}

[ritikrajdev]

CLA check is passing over here, you can proceed with reviewing the PR and further processes.

[emam07]

The two failing tests (TestShuffleNodes and TestMemberList_ProbeNode_Awareness_MissedNack) are pre-existing flaky tests unrelated to this PR.

This branch only modifies state.go — neither failing test exercises that code path:

• TestShuffleNodes (util_test.go) tests shuffleNodes() in util.go, which uses rand.Shuffle. With 8 elements there is a ~1/40320 probability the shuffled
  order matches the original — a known statistical flake.
• TestMemberList_ProbeNode_Awareness_MissedNack is a timing-sensitive test that already uses iretry.Run() and is known to be flaky on loaded CI runners.

Could you re-run the failed job? Happy to investigate further if it reproduces consistently.

[tgross]

@emam07 just a heads up that this is on my review queue. I'll try to get to it soon.