[CI] Try bumping job preparation limit for libc++ runners

We are seeing transient failures with workflow pods failing to initialize in
the default 10 minute timeout due to some transient image pull failures. Bump
the limit to 20 minutes to see if that helps things.