This needs careful deployment:
1. Disable puppet on high-traffic2 LVSes globally
2. Disable puppet on existing cache_maps hosts
3. Merge this change
4. Run puppet on all cache_upload (adds new LVS IPs)
5. Step through high-traffic2 LVS puppet-agent -> restart
  (which will rm maps cluster, add maps IPs to upload cluster)

note cache_maps hosts left puppet-disabled intentionally above

If something goes badly at this stage, we can revert similarly and
the cache_maps hosts remain in a ready state to take traffic back.

The revert process would look like:
1. Re-disable puppet on high-traffic2 LVS
2. Disable puppet on cache_upload
3. Merge revert of this change
4. Step through high-traffic2 LVS puppet-agent -> restart
  (which will re-create the maps cluster and move maps IPs to it)
5. Re-enable puppet on cache_upload and run agent (rm maps IPs)
6. Re-enable puppet on cache_maps (no-op)

Bug: T164608
Change-Id: I5c2ba9fd165fe344861fdf22e4914af163759728