Why are dependent sources never deactivated in application of this theorem?
When solving a network with multiple independent sources, we can use the superposition principle. This means deactivating all except one independent source one by one and solving the network for each case, then combining the results to get the solution for the network with all sources active. This is probably what your text is talking about when it talks about "deactivating" some sources.
You can't deactivate controlled sources when doing this process because a controlled source in a network with no independent sources wouldn't produce any output, so it can't be analyzed independently. The behavior of the controlled source has to be analyzed as it responds to each of the independent sources in turn.
Why is in the mentioned case "resistance combination is not applicable"?
In a network with controlled sources, you can't simply combine resistors like you do in a purely resistive network. Because there's more than just resistors in the network.
There are in fact special cases where a controlled source is hooked up to behave like a resistor, but in the general case, you have to treat them differently.