By nature, streams are interrupt- (aka event-) based. This makes it
unpractical to incorporate them into the main request's Promise chain.
Therefore, make the processing loop to happen asynchronously with
regards to the main Promise chain, i.e. the library's main entry point.
In the same vein, ensure that the main processing loop "takes a break"
every once in a while so that various interrupts and events can be
processed.

Also use cassandra-uuid instead of node-uuid for generating UUIDs, since
the former is a pure-JS implementation while the latter is a binary
module. Switch from UUID v4 to UUID v1 IDs because UUID v1 is used
throughout the system to generate the x-request-id and it also gives
us a notion of causality between different requests.

Bug: T199813