We're loading the full set of assets on every page, because we're loading all assets to
include them on every page. This diff changes the secondary asset query lazy, so that
only opening global search should incur that query cost.
We should also be rewriting the graphql server so that a single slow request
doesn't block all other requests