Configure dmsetup to fallback to managing device nodes without udevd

[stealthybox]

This patch frees up dependent interactions with udev for device node creation.
Ignite can now create boot devices in absence of a working udevd equivalent.
This makes ignite easier to run in docker containers and WSL2.

I wrote a similar implementation (no_udev_sync) which completely bypassed udev sync even when present.
It execs dmsetup info to retrieve block device paths and manages its own symlinks, but I found this implementation gets sort of coupled to /dev in the root mount namespace because dmsetup info will return the canonical device-path of /dev/dm-$i.
It's simple and possible to call mknod /dev/dm-$i -m600 253 $i and ensure the device node exists with no_udev_sync.

I decided to try using the built-in udev fallback from dmsetup which is what is implemented in this PR.

Note it's possible to actually get working behavior with older builds of ignite by setting DM_DISABLE_UDEV=1, but this requires the user to know about udev and potentially passthrough their env to sudo.

I found in dmsetup's changelog that the --verifyudev fallback actually used to be the default behavior.
I didn't look further, but I assume this was disabled, because it could be potentially harmful if you created devices during boot before udevd started up.
I figure that silently having udev rules not apply is fairly low-risk for ignite boot devices -- if people are using ignite during boot, they should be configuring their units at the proper run-level, but it is a consideration.
The trade-off is that if we accept this bit of informal fallback behavior, users don't have to configure dmsetup, and ignite can use dmsetup in containers without mounting /dev:/dev*.

Note: losetup does not seem to currently have any device node management.
In order for losetup to work in a container without mounting /dev:/dev you need to pre-create some loop device nodes in the container like so:

for i in {0..20}; do mknod /dev/loop$i -m600 b 7 $i; done

Other Changes:

• Fix DeactivateSnapshot bug which would cause cleanup failures
• Refactor ActivateSnapshot to return snapshotDevPath so it is just calculated once

[darkowlzz]

LGTM!
Great improvements. Will be fun to run in container 🙂