Major Changes in microvm.nix

Unreleased: `main` branch

Shell scripts to setup virtiofsd, interfaces, and PCI devices for pass-through have moved from systemd units in the host to the bin/ subdirectory of MicroVM packages. That means you can actually use them from the command-line if you would like to run with virtiofsd.

A breaking change is the move of the microvm.virtiosd.* options from the host to the microvm NixOS module. They were introduced only after 0.5.0.

New VMs still work on old hosts, but old VMs are missing the scripts. Switching your updated NixOS host will print a warning about which MicroVMs need updating.
bin/virtiofsd-run (microvm-virtiofsd@.service) now starts the multiple virtiofsd instances through supervisord.
The microvm module allows configuration of microvm.virtiofsd.group and microvm.virtiofsd.inodeFileHandles and microvm.virtiofsd.threadPoolSize now.
Add the alioth VMM
Fixes for the stratovirt VMM
New volume image files will be created with truncate instead of fallocate, saving disk space.
Building the microvm.storeDisk is now faster as the system closure is no longer first copied into the build directory. A mount namespace is setup with bubblewrap instead. mkfs.erofs can now run multi-threaded.

tap interfaces are now multi-queue when running with more than one VCPU. Update your host!
The host module enables Kernel Samepage Merging by default.
qemu can run non-native systems by using its Tiny Code Generator instead of KVM.
SSH deployment scripts are added as config.microvm.deploy.rebuild
qemu defaults to the microvm machine model now as it supports PCI, USB, and ACPI by now. Set microvm.qemu.machine = "q35" if this breaks for you.
The NixOS hardened profile can be used by falling back to squashfs.
Runners execute the hypervisor with a process name of microvm@$NAME
We no longer let environment.noXlibs default to true
Breaking: the microvm user is no longer in the disk group for security reasons. Add users.users.microvm.extraGroups = [ "disk" ] to your config to restore the old behavior.

cloud-hypervisor replaces rust-hypervisor-firmware with direct kernel+initramfs loading.
The microvm module now optimizes the NixOS configuration for size.
crosvm now supports macvtap interfaces.
The option microvm.qemu.bios has been dropped again for simplicity reasons.

qemu boots fast with the shipped SeaBIOS if after both SATA and the network interface option ROM (iPXE) have been disabled.
microvm.kernelParams always copy boot.kernelParams
firecracker is no longer launched through firectl.
Networking example documentation has been split into multiple scenarios.
Vsock support has been added for Hypervisors that connect them to the Linux host's AF_VSOCK: qemu, crosvm, and kvmtool.
Our packages and overlay include the unstable version of waypipe, featuring Vsock support.
Add support for the old command-line parameter syntax that returned with cloud-hypervisor 36.0.

Stop building a custom kernel by booting the NixOS kernel with an initrd.
New Hypervisor: stratovirt by Huawei
Support fully declarative MicroVMs that are part of the host's NixOS configuration. No Flakes required!
We use squashfs-tools-ng now.
The microvm-console script has been removed because pty console setup was too cumbersome to maintain across all hypervisors.
microvm.storeDiskType defaults to "erofs" now for higher runtime performance.