drop dracut for initramfs generation

[ddiss]

Dracut is useful for providing a bunch of system image "recipes" in the form of dracut-modules, as well as performing a few early-boot tasks before rapido autorun. I initially expected to write and use dracut-modules for rapido, but having dracut source internal (system) and external (rapido) modules simultaneously proved difficult.
Nowadays we tend to only use the base and systemd modules, with one exception for multipath. As such I think we could consider dropping dracut and instead using the kernel's gen_init_cpio binary.

I think the main benefits of this change would be:

• portability: dracut isn't used for boot-time initramfs generation on many distros. some are also moving from dracut to mkosi.
• performance: dracut is quite a fork-bomb, and stages all files before generation to accommodate GNU cpio's silly source-is-dest behavior
• simplicity: having full control of init should reduce the amount of regressions caused by upstream dracut changes

I think the main tasks to move things forward here are:

• add logic to determine kernel dependencies via modules.dep
• parse ldd output to gather shared object dependencies
• convert existing dracut invocations to gen_init_cpio archive manifest files

[ddiss]

FWIW another option I'd considered is going down the rust path and reusing my dracut-cpio archiver. Only problem is that it follows the same cpio source-is-dest usage behavior.

[ddiss]

FWIW another option I'd considered is going down the rust path and reusing my dracut-cpio archiver.

I ended up going down this route, as there's a rust-elf crate which handles recursive dependency checks very well. I plan on submitting the WIP changes soon, with associated rust-based vm and init binaries.