twister: optimize file scanning

We have been scanning code for tests for every scenario defined in the
yaml file. This needs to only be parsed once and not more. We are
dealing with the same files for each scenario.

Signed-off-by: Anas Nashif <anas.nashif@intel.com>