Add tests for LiveNodes refresh, fallback, and concurrent startup behavior

[m-szymon]

Current tests do not cover several LiveNodes refresh and startup edge cases in the update_live_nodes path and discovery startup flow.

Include:

• empty /localnodes with no fallback leaves the current live node list unchanged
• failed /localnodes request retries remaining candidates in the same scope
• seed nodes are retried when all current live nodes fail
• total refresh failure does not wipe out the previous known-good node list
• concurrent first access starts discovery exactly once

Expected outcome:

• refresh failure and fallback behavior is pinned directly
• existing live node state is preserved on empty or failed refreshes
• seed fallback works as intended
• concurrent startup does not spawn duplicate discovery tasks

Part of #41.