[Feat] Configure Netdata's pgbouncer collector when both are installed

[violuke]

Problem

The role installs Netdata on the Postgres nodes (netdata_install defaults to true) and installs pgbouncer on
those same hosts, but nothing sets up Netdata's pgbouncer collector. The Postgres collector finds itself
through auto-detection so you get postgres.* charts for free, but pgbouncer stays invisible. Relates to #249

That gap is awkward in practice, because pgbouncer is the thing that actually governs connection capacity. We
were recently digging into connection behaviour on a cluster deployed with this role and had to SSH in and run
SHOW POOLS by hand to get numbers a collector would have been graphing all along. maxwait creeping above zero is exactly what you'd want a chart and an alarm on, and right now there's nothing.

Description

The netdata role runs the kickstart installer and templates netdata.conf. It doesn't write anything under
go.d/, so there's no existing place to hook this in.

Rough shape of what I had in mind:

netdata_pgbouncer_collector: true
netdata_pgbouncer_stats_user: netdata

templating /etc/netdata/go.d/pgbouncer.conf:

jobs:
  - name: pgbouncer
    dsn: 'postgres://{{ netdata_pgbouncer_stats_user }}:{{ netdata_pgbouncer_stats_password }}@/pgbouncer?host=/var/run/pgbouncer&port={{ pgbouncer_listen_port }}'

The unix socket seems the right default for two reasons: it sidesteps client_tls_sslmode = require, which
only applies to TCP clients, and it keeps the credential off the network. The user would need to be in
pgbouncer_stats_users, which is currently just patroni_superuser_username, so that would want extending
as well.

Worth flagging that I haven't tested any of this. It's the shape I'd expect rather than a working patch. The
bit I'm least sure about is pgbouncer_processes > 1, where pgbouncer.ini.j2 gives each process its own
socket directory with an index suffix, so a single job would only ever see the first one.

Thanks for considering this and for autobase itself, it's great!

Importance

really want

Proposed implementation

No response

[vitabaks]

@violuke Feel free to offer PR.

[vitabaks]

so a single job would only ever see the first one.

Yes, this is a disadvantage of the so_reuseport approach.