Sync process recipes to avoid fetching unnecessary dependencies

[kylewlacy]

See this Zulip discussion for some background context: https://brioche.zulipchat.com/#narrow/stream/440653-general/topic/Why.20so.20slow.3F/near/442990093

When running the snippet to download "hello world" on the brioche.dev landing page (brioche run -r hello_world), it would take a long time to run, because it would end up downloading all of the inputs used to build hello_world (even though these were never used at runtime). This was caused by a split between process and complete_process recipes (complete_process is an implementation detail so that even if the input recipes of a process recipe change, it can still be a cache hit as long as those recipes bake to the same artifacts of another process recipe).

Basically, we were only caching the inner complete_process recipe, which in turn meant that clients would end up trying to bake all of the inputs of the outer process recipe, which is why all the dependencies would be fetched.

This PR changes it so that both process and complete_process get synced to the registry from now on. This should make it so the next time the brioche-packages repo gets synced, it should automatically cache the missing process recipes. However, this won't retroactively fix existing clients, since they would never even check the cache for process recipes.