Rather than performing this step in the root CMakeLists, as we previously did
in the "server_layout" target. This effectively replaces the "server_layout" target.
Reorganizing the sparkd proj with stuttering dirs broke the build because
the generated protos were being dumped into a dir with the same name
as the "sparkd" executable.
So we can be more precise when generating protobufs. For example, now we
can specify different include dirs when generating code for different
group of proto messages - an obvious improvement over the previous
system, which relied on a hardcoded superset of include dirs that would
support every proto message in our code base.
CMake does this automatically now (as of version 3.20). Also, for some
reason unbeknownst to me, cmake doesn't clean the generated proto files
when they are explicitly marked as such.