Enable shallow checkout of recursive submodules

[andrjohns]

The current handling of submodules results in a shallow checkout only for the toplevel submodule, while the recursive submodules are cloned in full.

This reduces the install time for StanHeaders from ~2mins down to ~30secs:

system.time({remotes::install_github("stan-dev/rstan/StanHeaders", force=TRUE)})

CRAN:

 user system elapsed 72.187 13.088 131.361 

This PR:

* DONE (StanHeaders)
user system elapsed 33.923 4.648 20.171