Agree with all of your points. The thing that perhaps annoys me the most is that people seem to assume that nosql is magically fast and that sql is slow... Sure, a key-value pair lookup runs like shit off a shovel, but as soon as you want to run anything more complex than that, it's likely faster with a relational database and a few indexes...
Usually the same people that think running an application on a cloud platform magically makes it fast, when in reality a VPS would be cheaper and faster.
NoSQL is not a well-defined term - there are huge differences between various solutions all commonly called "NoSQL", so the comparison to NoSQL is bogus. There are different NoSQL things designed with different use-cases in mind. On one side you'll have things like MongoDB, which is easy to set up, but doesn't really scale once your dataset grows out of memory, on the other side things like Cassandra which are unmatched by any RDBMS in terms of performance, scalability and availability on real-time, transactional workloads. Of course, your RDBMS of choice might be better at complex JOINs than Cassandra (which doesn't have JOINS at all, btw), but will it stand the chance at performace competition with Apache Spark or Hadoop? I don't think so.
Usually the same people that think running an application on a cloud platform magically makes it fast, when in reality a VPS would be cheaper and faster.