Simply one is bi-variate and the other is multivariate.

one versus another variable, and the other versus many variables

do both of these test short term causality only ? Can we apply both to non co integrated data ?

depends on whether or not you are using stationary data or non stationary variables with an existing error correction model.

yes you can apply them to non co integrated data, to stationary data like first difference or log returns for example

Yes, you can apply only the variables are non cointegrated