Optimize performance on DiscoveryNodeManager

[guhanjie]

Description

Avoid unnecessary nested loops O(N*N) on DiscoveryNodeManager#refreshNodesInternal

Additional context and related issues

It does nested loop O(N*N) when services filter out failure nodes on DiscoveryNodeManager#refreshNodesInternal, it's no big deal on small clusters with few nodes, however it consumes quite a lot of cpu cycles on large cluster with over 1000 nodes, that will do 1,000,000 cycles per round of refreshNodeInternal. According to acutal cpu profiling, we found jvm does not optimize the code automatically, so this commit optimize it and avoid nested loop.

Release notes

( ) This is not user-visible or docs only and no release notes are required.
( ) Release notes are required, please propose a release note for me.
(x) Release notes are required, with the following suggested text:

# General
* Reduce unnecessary nested loops overhead on DiscoveryNodeManager#refreshNodesInternal.
   This can avoid unnecessary nested loops while services filtering out failure nodes on DiscoveryNodeManager.

[guhanjie]

@findepi please merge the approved PR

[wendigo]

@guhanjie please improve the commit message so it renders in a github without ellipsis

[guhanjie]

@guhanjie please improve the commit message so it renders in a github without ellipsis

PR title updated: Optimize performance on DiscoveryNodeManager
Do you mean PR title or commit message, which one will introduced into release notes?
the code has been merged, i cannot modify any more..