Move away from Travis

[waldoj]

Travis is now Actually Bad. Also, I cannot get clean builds anymore, due to goofy errors that I gather are becoming more common with Travis. I haven't been able to deploy for months. Instead of fighting a series of little fires, better to complete the frustrating task of rewriting the Travis script for Circle or GitHub Actions.

• get a version of the build/deploy config so that it can run without errors
• add every secret from Travis into GitHub
• restore testing
• enable running builds (not not deploys) on PRs
• test and debug cautiously, reviewing all GitHub Actions output
• permit the script to function on the deploy branch
• add status badge to README
• repeat all of this for rs-api and rs-machine

[waldoj]

This Travis -> GitHub Actions converter should make this a lot easier. It's far from complete, in what it can auto-convert, but it's a start.

[waldoj]

Well, that was a start, but not enough of one. So I'm starting from scratch. I'm leaning on these two resources: Migrate From Travis CI to GitHub Actions and Amazon EC2 Deployment: Complete CI/CD Pipeline using GitHub Actions and AWS CodeDeploy.

[waldoj]

Well, I got a successful deploy...but it deployed the staging site to deploy, which...🤯

[waldoj]

I've got two problems that I think are the same problem.

In build, we move the environment-specific appspec to appspec.yml, so that CodeDeploy will use that...but that file is missing from the actual deploy bundle sent to AWS, so the deploy fails.

And in deploy, we start with uses: actions/checkout@v2, and the log shows that, indeed, the repo is being checked out. Which is sensible enough, but that means that none of the steps in the build process are persisting to the deploy step, which is, of course, bad.

I think the solution to both of these things is to replace that pasted-in step with something that will instead draw on the output of the build step.