Summary

• fix BalancedPool weighted round-robin recovery after outages when all upstream weights reach the minimum value
• avoid mutating scheduler state during dispatcher availability probes by adding a non-mutating peek mode to kGetDispatcher callers
• add a regression test that reproduces the 2-upstream outage/recovery scenario from BalancedPool deteriorates to a single point of failure after outages #4839

Details

When two upstreams both degrade to weight 1, BalancedPool could repeatedly select the same failed upstream and never probe the healthy one during recovery.

This was caused by two interactions:

1. stateful dispatcher selection was being invoked for a "can dispatch?" probe in PoolBase, advancing scheduler state unexpectedly
2. weighted round-robin weight reset used maxWeightPerServer instead of the current max available upstream weight

Testing

• npx borp -p "test/node-test/balanced-pool.js"
• npx borp -p "test/pool.js" -p "test/round-robin-pool.js"
• npm run lint

Closes: #4839