radv: Prepare gang submissions to work with the SDMA + ACE use case.

Based on !25770 (merged)

The current gang submission functionality in RADV assumes that it is only used by task shaders. This MR prepares the functionality necessary to support the use case for SDMA + ACE:

• Simplify the initialization of gang follower CS and semaphores to make it less tedious to call. It will be called from multiple other places in the future.
• Add possibility to for gang leader to wait for the follower. This was unnecessary for task shaders because DISPATCH_TASKMESH_GFX implicitly waits for DISPATCH_TASKMESH_*_ACE but it will be needed in the future.
• Remove the assumption that the gang leader must be GFX and the follower is only needed for task shaders.