Support asynchronous kernels

[calvis]

In particular, this means adding an async-kernel form. This may not be necessary. However, the language has enough information that we can probably generate asynchronous code for kernels by default, which means we can automatically overlap CPU and GPU computation.

[eholk]

We have a test case with some example syntax here: https://github.com/eholk/harlan/blob/master/test/async-kernel.kfc