* Always dump the current output/result

This fixes an issue that was introduced in cloudtools#467. The original
functionality was to always dump the current output to `.json-result`,
that way it was easy to build & compare the output.

With the current implementation (from cloudtools#467) we raise an exception if the
old results aren't available, but don't create the new results.  This is
an issue whenever creating a new test.

* Update CHANGELOG

This changes the internals of stacker plan's to first generate a
Directed Acyclic Graph, and uses this as a means of walking the graph.

Generate a DAG internally

Remove DescribeStacks call from prepare_stack_for_update

Simplified logging

* add raw json/yaml template support

Fixes cloudtools#444

* fix diff; fix transform; remove build action hack

* add raw template tests

* additional updates for raw template support

* update diff action to support regular & raw blueprints without
  conditions
* fix validation & test of class / template path use in Config
* move raw blueprint specific function from util to blueprint module
* cleanup class selection in stack.py

* Handle mutual exclusion validation better

* Add functional test for template_path

* Remove dependency on troposphere template

* Make RawTemplateBlueprint inherit from object.

* cleanup pylint/pydocstyle messages in raw blueprint

Color failed stacks as red

Parallel graph walk

Ref!/Sub!/etc will no longer fail to load

…udtools#638)

* allow raw cfn templates to be loaded from remote package_sources

Fixes cloudtools#628

* add test for new get_template_path function

getmoto/moto#1793

Until we either move entirely off moto to botocore.Stubber, this is the
easiest workaround.

…udtools#649)

This could speed up stacker by many seconds based on the profiling work
@ejholmes performed in cloudtools#630

	modified:   stacker/providers/aws/default.py

Weird type issue in setup.py

pinning needed till getmoto/moto#1793 is resolved

Specify version dependencies as documented in
https://docs.python.org/3.6/distutils/setupscript.html#relationships-between-distributions-and-packages.
Using the '~=' format causes OS packages built with fpm to have a wrong
dependency.

(note the '=2.0' instead of '~=2.0')
The following packages have unmet dependencies:
 python3-stacker : Depends: python3-gitpython (= 2.0) but 2.1.11-15 is 
to be installed
                   Depends: python3-schematics (= 2.0.1) but 2.0.1-15 is 
to be installed
                   Depends: python3-dateutil (= 2.0) but 2.7.3-15 is to 
be installed

…cket.yaml (cloudtools#656)

Instead of implicitly getting an ASCII bytestring,
Explicitly get an UTF-8 bytestring.

Don't try to decode anything afterwards.

Resolves this error:

```
  'ascii' codec can't decode byte 0xcb in position 5: ordinal not in range(128))
```

	modified:   stacker/lookups/handlers/kms.py
	modified:   stacker/tests/lookups/handlers/test_kms.py

…unicode

Fix KMS lookup for Python2

…loudtools#657)

create change set doesn't allow a `stack_policy` argument like
create/update stack calls.  This was handled in the interactive version,
but not in the non-interactive version.

Fixes cloudtools#650

…dtools#658)

PR cloudtools#644 didn't account for empty string parameters

Hoping this helps with what appears to be harsh throttling issues.
Long-term: move this to AWS codebuild

This adds "targets" to stacker, which are just named nodes in the graph
that specify dependencies.

Why would you want this? Primarily just as an easy way to specify a
logical grouping of stacks that you can use with the --targets flag
(e.g. maybe you want a target for "databases" or "apps", etc).
Essentially, these are no different than stacks, except they don't incur
any AWS API calls, and don't take up a precious CloudFormation stack
spot.

If you're familiar with targets in systemd, the concept is exactly the
same; a node in the graph that does nothing.

Targets

Locked stacks have no dependencies

…file

Set default AWS profile earlier

* Get rid of recursion for tail retries

Also, extend the way we retry/timeout, that should work around cloudtools#515.

Not sure of a great way to test this unfortunately.

* Add some tests

…ls#670)