Refactoring + optimization

[greatroar]

Here's some refactoring of the common code for dealing with rolling windows.

The first patch factors out the WriteWindow methods to a function in an internal package. The WriteWindow methods are now so small that they get inlined, so callers still do a single function call (not counting the Write calls).

The second factors out the re-arranging, while changing the algorithm to operate in-place (no allocation).