zaml: rework strings for correctness and speed

This changes the implementation of string output in the ZAML encoder. On my
test that delivers around a fifty percent reduction in output time, from 4.11
to 1.89 seconds on my test of ~ 35,000 output strings.

It also adds a bunch of tests to validate the correctness of binary output
handling, and some test inputs derived from various security related UTF-8
encoding problems to validate that (sadly) we pass them through as
binary data.

Optimally we would treat binary and text input as distinct in the parser and
compiler of Puppet, then only allow legal UTF-8 in strings - but that is not
the world in which we live.

It finally fixes a bug in the binary output along the way - which, thankfully,
nobody actually seems to have run into since I introduced it in ea0dd1483fc2.

Because the Ruby BASE64 encoder will split lines at the 60 character mark, a
long binary input would end up with embedded newlines in places that were not
legal for YAML.

Instead of just emitting that, we run through the string encoder over the
transformed data, which leads to a solid output format that works entirely
as expected.

Signed-off-by: Daniel Pittman <daniel@puppetlabs.com>

Conflicts:
lib/puppet/util/zaml.rb
spec/unit/util/zaml_spec.rb

Signed-off-by: Daniel Pittman <daniel@puppetlabs.com>