I don’t use the Arr suite or Qbit (nor do I really torrent that much) so I can’t speak on the second part, but for scaling I use Ceph. I currently have about 95 TiB across 3 machines and based on my experience, scaling it up further (ie adding more to a machine or adding new machines) is fairly straightforward and relatively simple. That said, I have my cluster set up to make 1 copy of data across each machine and have a few TiB reserved for metadata so I only have about 29 TiB for unique object storage but that kind of setup isn’t strictly necessary. You could set up your own cluster such that there’s no redundancy and utilize most of your available storage for unique objects (a relatively small portion of it will still need to go to metadata if you want to set up a Ceph filesystem, though, but that also isn’t necessary).

Gentoo for most of my personal machines. I currently have about 12 that I use actively (bare metal + virtual).

(Among other things,) I currently use Ceph across 3 servers for storage; Buildah/Podman/Skopeo, LXD, and Libvirt for virtualization; Git for versioning/a simple way to keep certain things in sync; and Saltstack to automate updates.

I have a dedicated virtual machine for building software packages which shares those built packages (currently via Virtiofs) with a LXD instance that exposes them over HTTP for my other machines to download so software only needs to be built/packaged once.

Personally I like and have been using beancount with my own instance of fava. It’s text based and may take a bit of getting used to coming from quickbooks, though.