Functions for setting up computational pipelines via data streams.

NOTE: This library is experimental. It may change significantly
with future release.

This library defines:
- an abstract stream type, whose interface consists of the
multimethod stream-next
- a macro for implementing streams
- implementations of stream for
1) Clojure sequences, and vectors
2) nil, representing an empty stream
- tools for writing stream transformers, including the
monad stream-m
- various utility functions for working with streams

Streams are building blocks in the construction of computational
pipelines. A stream is represented by its current state plus
a function that takes a stream state and obtains the next item
in the stream as well as the new stream state. The state is
implemented as a Java class or a Clojure type (as defined by the
function clojure.core/type), and the function is provided as an
implementation of the multimethod stream-next for this class or type.

While setting up pipelines using this mechanism is somewhat more
cumbersome than using Clojure's lazy seq mechanisms, there are a
few advantages:
- The state of a stream can be stored in any Clojure data structure,
and the stream can be re-generated from it any number of times.
Any number of states can be stored this way.
- The elements of the stream are never cached, so keeping a reference
to a stream state does not incur an uncontrollable memory penalty.

Note that the stream mechanism is thread-safe as long as the
concrete stream implementations do not use any mutable state.

Stream transformers take any number of input streams and produce one
output stream. They are typically written using the stream-m
monad. In the definition of a stream transformer, (pick s) returns
the next value of stream argument s, whereas pick-all returns the
next value of all stream arguments in the form of a vector.

Vars in