When writing a diff to a file section, we always ensure it ends in a

newline. There's some work that's done to ensure content and newlines

are in the right encoding and the BOM stripped, and this didn't work

quite right when working with a specified newline type, as opposed to

guessing a type.
The result was that, for multi-byte-encoded diffs, we could append a

newline with a BOM, breaking the content.
This logic has been cleaned up to ensure we always encode and strip BOMs

at the right time.
For additional reliability and performance, we also no longer split

content into lines unnecessarily. If we're not indenting, we just return

the content as-is instead of splitting into lines and writing them to a

stream first.

Unit tests pass on Python 2 and 3.