[PDS-332497] getNamespaces() on a busy timelock server is costly

[jeremyk-91]

The way the timelock server does the scan to find the list of namespaces could read through more data than we expected. We do a SELECT DISTINCT (namespace) FROM paxosLog where there could be O(200) namespaces but the log itself can have millions of entries - although there is an index, it still needs to page through 100000s of copies of the same string there.

The solution here is probably to manually maintain a table of the namespaces and use database triggers or something to keep that in sync - basically a manual index - though doing a suitable migration requires a bit of thought!

This endpoint is only really used by timelock-migrate, and it's not usually so bad of a problem in that while it's expensive we can run it on one node at a time (so this is losing HA, not generally causing outages)... unless we need it when the cluster is degraded, as in the ticket.