Fetching CoreDMA handles in device_setup()

[pmldrmota]

In some cases, it would be mildly useful to have a second device_setup(), e.g. device_setup_postaction() that visits all subfragments once again before run_once().
The primary use case for this would be to acquire DMA sequence handles in different fragments after they were recorded in device_setup(). [The core_dma's epoch counter prevents both these steps to be done in each fragment subsequently.]

[dnadlinger]

I understand the use case, but adding a second stage doing exactly the same seems a bit arbitrary/overly specific in terms of ndscan design. I wonder whether this would be best implemented in a user-specific subclass like ReadoutExp (Oxford-internal), given that the entire fragment tree is known at host execution time anyway, where the walking can be done even without ndscan support.

Really, though, the best option might just be to overhaul ARTIQ DMA buffer management to avoid this altogether and keep handles valid unless the same sequence is modified. The more I think about this, the less I am sure why this isn't how it is done already.