Replace autopilot zero down time deploys with own implementation.

[dam5s]

The goal of this change is to ensure we always use cf push as implemented in the cf CLI.

This fixes issues with zero downtime deploys of big applications (timing out during diff evaluation), and zero downtime deploys of docker based applications.

I also added a test and fix for https://github.com/concourse/cf-resource/pull/71

[making]

@vito Could you please prioritize this?

Since autopilot is no longer updated and it uses vendored old cli, recent changes in manifest.yml are not supported.

For example, buildpacks field is not supported (see the error message below) while buildpack field has already been deprecated. Currently using cf-resource forces us to keep on using legacy fields.

error: Error reading manifest file:
The following manifest fields cannot be used with legacy push: buildpacks

[vito]

Sorry! I'll take a look at this soon.

[dam5s]

Just rebased and resolved conflicts.

[dam5s]

There are three reasons I replaced Autopilot:

• It was written as a CF plugin and did not support all the flags that cf push did.
• It was invoking a cf push function instead of shelling out which seemed to make it fail when pushing big apps. But it could be due to the fact it was depending on an older version of the CLI.
• It is no longer maintained.

I did not cover all the cases because they didn't seem to cover any cases that would occur in the context of a CI pipeline.

So far I have been using this altered version of the cf-resource for a few pipelines and it seems to cover all the cases that pipelines seem to have... but I could be wrong.

It now does support all the cf push options which is why I did the change in the first place.

Happy to re-implement the missing edge cases if it happens to cover cases that would be useful to the cf-resource.

[vito]

Sounds good. :+1: