feat(roles): run the database dumps with low CPU and IO priority

[markuslf]

Summary

• The daily dump services of gitlab_ce, influxdb, mariadb_server, mongodb and postgresql_server run with Nice=19, IOSchedulingClass=best-effort, IOSchedulingPriority=7, CPUWeight=20 and IOWeight=20, so the dump no longer slows down the database. IOWeight only takes effect with the BFQ I/O scheduler. The READMEs got a section "How the Role Behaves".
• MongoDB and MariaDB deliberately do not use the idle classes: the MongoDB dump holds fsyncLock, and mydumper holds LOCK TABLES ... READ LOCAL on non-InnoDB tables until each table is dumped.
• mariadb-dump only passes the mydumper options the installed version knows (--dirty, --trx-tables=0, --threads=0), so the dump works with mydumper 0.10.x and 1.0.x.
• The daemon-reload handlers already on main cover the changed unit files, so the tasks stay untouched.

Test plan

Before rebasing onto main (the task files were identical apart from a daemon_reload that the handlers now replace): mariadb, mongodb and influxdb on Rocky 8/9/10, Debian 12/13 and Ubuntu 22.04/24.04/26.04, PostgreSQL on RHEL 9/10, Debian 12/13 and Ubuntu 22.04/24.04/26.04. Where the role ran, the units carried the new values, the dump Result=success and the second run changed=0. CPUWeight is mapped to cpu.shares on Rocky 8 (cgroup v1).

Not testable: gitlab_ce (only systemd-analyze verify), MongoDB on Debian 13, Ubuntu 26.04 (no upstream packages), PostgreSQL on Rocky 8 (PGDG key), MariaDB dump on Debian 13 (no mydumper, see the mydumper repository PR).