all: ensure we can run genuine unit tests from Librarian

Previously we've been running `go test ./... -short` in Librarian automation.

This fails for storage, spanner and bigquery (at least) because the tests appear to expect to be running in an environment where there's a project available (e.g. via a Compute metadata server).

The Spanner tests may not have this problem, but fail with a timeout after 10 minutes. (I haven't checked whether this is configurable, so that may be fixable.)

I'm currently creating a PR to remove `go test` from the container (librariangen) but it would be better if we had a convention of "tests that are run by `-short` (or some other flag; I don't mind what) do not require any external resources or environment".

Assigning to Seth to work out where to go from here.