Saving snapshots to disk should not delay collection

Due to recent GPFS issues saving snapshots to disk took >10s and so timed out Hyperion. To get round potential timeouts like this we should:

1. Take the snapshot and hold it in memory
2. Continue collection once this is done
3. Asynchronously write the snapshot to disk - including grid snapshots
4. It's ok for 3 to take a long time (~1/2 mins)

We still need to be able to take another snapshot whilst 3 is taking place without affecting the saving of the first images.

It may make sense to move to doing this in callbacks rather than the ophyd device.