ci: Use AUDIO and INFERENCE options with ML applications only

For non-ML applications there is no need to build and run for
AUDIO and INFERENCE variants as these configurations should
be relevant for ML applications only. Non-ML applications would
use the default values only. Thus, separating non-ML
applications pipeline configurations from the ML applications
pipeline configurations.

Signed-off-by: Ahmed Ismail <Ahmed.Ismail@arm.com>