I think the conversation really comes down to 2-3 factors: Price, Power, Footprint. Anyway, back with spinning disk, you'd look at performance as well as loss to root-aggregate/etc. With SSDs, the common agreement seems to be that even a single shelf of 15TB drives can exceed the performance capacity of the controllers - so going to more disks (as you might with SAS or SATA to get more IOPs) doesn't really play a major factor. With ADPv2, loss to root-aggregate becomes almost a non-consideration with larger drives.
So then, which configuration is cheaper? Do you have power/heat constraints in the data center where these will live (larger drives = less power/heat)? Do you have limited rack space (again, bigger drives = less footprint)? The AFF A700 spec sheet shows it shipping with the following drives (initial configuration):
Configured with new systems:
Storage shelf: DS224C; 2U 24 drive, 12Gbps SAS
SSDs: 15.3TB, 7.6TB, 3.8TB, and 960GB
We ended up going the 15TB route because of rack space considerations. Had the price been super compelling to go smaller we might have done that (its easier to squeeze more space than squeeze more $$ sometimes) but it didn't end up changing pricing too much for us...
The whole "single shelf" thing is sort of an interesting situation. If you've got two shelves, odds are you're going to be spanning your raid-groups across disks in both shelves (otherwise you're building some small-ish aggregates) so losing one-of-two shelves in that scenario still takes your system offline. There is a veritable ton of redundancy/etc built into NetApp shelves so the likelihood of you losing a shelf seems pretty unlikely. We had a bunch of dual-power supply SATA shelves and everytime our facility folks did power work on one leg of power I was always paranoid that we'd lose the "hot" shelf power-supply during the maintenance window - never happened, but honestly I think that's the most likely failure scenario.
In the end, if you're super-worried about data availability and want to design around a single shelf failure, you probably ought to be looking at mirroring your aggregates (SyncMirror) across shelves in a way such that the raid-groups don't span the shelves (i.e. each shelf can act independently). But you'll need to worry about your root aggregates in the same manner - i.e. making sure that a single shelf can't take down the whole system. That said, we've had multiple NetApp systems go down "dirty" (thermal event in one situation, complete power loss in another, etc) and they've never failed to come back up, replay NVRAM and just keep on trucking with no data loss/corruption.
The DS224C is the new 12Gbps SAS-attached shelf "optimized" for SSD throughput. The DS2246 is currently supported but is down-revved to 6Gbps. Again, there are a ton of variables that could represent the limiting factor in performance on an AFF system, but there's no sense in starting off on ~7yo SAS technology with your new/speedy SSD disks. Definitely work with your VAR and sales engineer to spec out the "right" system for your situation...