It's promising, at least. I think a system like this would be the most useful if you're able to layer a query language on top of it, and apparently even FoundationDB (which had an SQL layer on top of a K/V store) didn't perform all that well. The Cockroach guys claim to have found a performant way of doing SQL on top of K/V, if I remember correctly, but we'll see. But having transactions in a distributed K/V store is certainly useful.
It does, but unless you know it is on top of a distributed KV store and design your schema and queries around that you will suffer horrible performance issues. Also, F1, similar to FDBs implementation, forces you to model your databases differently than you would in a "normal" oltp relational database.
Almost every single host SQL solution I know of is built on a B-Tree for at least some of its properties. A B-Tree is nothing more than a lexicographically sorted key-value store.
Consus is a lexicographically sorted key-value store.
Where you will hit performance bottlenecks is when your query planner/executer makes traditional assumptions about performance that do not hold across machines.
A good SQL implementation on top of a kvs should not force you to accommodate it any more than your current "normal" databases subconsciously train you to accommodate your disks (hint: they do).
I would be very surprised if anyone is still running their primary data store on "disks" that cares about performance. BTW, look at the design of F1 and you will see that they do object like hierarchical storage of related tables to ensure that they don't have to go cross partition for queries.
