fix: detangle code and unmount old vols on start

[olevski]

A few things change here:

• I added a function that will unmount volumes from the code rather than doing it with a shell script from a lifecycle hook in k8s
• CSI drivers have 2 components - a node service and a controller service. The node service has to run on every node and in our case has to run rclone. The controller service is much simpler and there should be only 1. Also only the node service can unmount the volumes at startup. Before this change, rclone ran on the controller and the node service. But now it only runs where it is needed.
• We build a single executable that can run both components. This is good. But I added a cmd line args to make it so that only the needed stuff runs on the node or controller services. Not everything in all cases.
• Weird things were pinned in the go.mod file. I suspect this had to do with the nix setup. It took me several hours to figure out why my packages were not updating.

I tested manually in a kind cluster that I can restart the node service while there are actively mounted volumes. And the volumes were correctly unmounted and the node service started back up.

[Panaetius]

Nice changes! Just two minor changes, not really related to the logic itself.