LGTM!

There are now three manual "-virtfs" arguments to qemu.  It's starting to irk me
a little.  It might make sense to generate those from %linux-vm-filesystems
in order to not have so much duplication and places where things can go wrong
when maintaining.

If you'd like, could you maybe pass the %linux-vm-filesystems to
load-in-linux-vm and then automatically calculate those -virtfs entries?

(In a future patchset - I don't wanna hold up this one)

There's already common-qemu-options which would support <file-system-mapping>,
and %store-mapping, so actually it would make sense to use those (probably easiest
to move build/vm.scm's load-in-linux-vm into system/vm.scm
expression->derivation-in-linux-vm so the qemu stuff isn't scattered all over
the place - right now, half is in gnu/system/vm.scm and a mostly redundant half
is in gnu/build/vm.scm ... not ideal).

There's a "mapping->file-system" helper, so it might make sense to use
<file-system-mapping> instead of <file-system> in %linux-vm-filesystems.

So all in all:

* Modify %linux-vm-filesystems to be file-system-mappings instead of file-systems.
Also modify the place it is used.
* Move load-in-linux-vm to gnu/system/vm.scm (maybe inline into caller).
* Modify load-in-linux-vm to use common-qemu-options.
Notice that the "-append" option is passed twice.  Bug... <-- I just fixed that in master

If you want I can do it.