Gets the API service working in ECS. Online SP now works in staging!

[willroberts]

This gets practice games working online! You can create an account and play a practice game at https://staging.duelyst.org.

This is another massive PR, with the following changes:

App Changes:

• Passes the game_type into NetworkManager's connect() function, which will make it easier for us to differentiate between Game & SP services running on the same domain.
• Mildly refactors the WebSocket connection logic in app/sdk/networkManager.coffee to split out the determination & validation of the WebSocket URL from the connection logic. Also enables retries in local development to slightly reduce code duplication.

Server Changes:

• Adds support for configuring Firebase authentication via new environment variables: FIREBASE_PROJECT_ID, FIREBASE_CLIENT_EMAIL, and FIREBASE_PRIVATE_KEY. These are used in staging to supply credentials from AWS SSM Parameter Store.
• Cleans up the process of downloading index.html and register.html from S3 in API, with more logs and more reliability when constructing URLs/paths (which previously used string addition).
• Cleans up isDevelopment() / isStaging() / isProduction() logic to simply differentiate development vs. not development, resulting in more consistency between staging and production modes.
• Disables email sending for all use cases, since we don't have a working Mandrill config / subscription currently.

Build Changes:

• Adds Dockerfiles for API & migrations.
• Cleans up the base Dockerfile, adding some Python/GCC/G++ dependencies for bcrypt in API/Worker.
• Adds a helper script for building the staging frontend code (scripts/build_staging_app.sh).

Infrastructure Changes:

• Gets the API service in ECS working, including healthchecks from the LB.
• Adds a database migrations service in ECS which can be selectively enabled to run migrations.
• Sets container memory allocation to 350MB. I was hoping for 512MB, but AWS gives us 936MB instead of 1024MB, and underlying processes like dockerd and containerd consume 150MB+ already. :(
• Adds more environment variables and secrets to ECS services, such as setting NODE_ENV, configuring Firebase, and enabling all cards in staging.
• Adds redirects to CloudFront+S3 from the LB for static assets, avoiding API entirely to reduce load.
• Fixes CORS configuration for CloudFront and S3.
• Loosens the requirements of open ports and LB target groups for ECS services, for things like migrations which don't need ingress traffic.
• Adds support for the CloudWatch Agent (which enables Disk/Memory metrics in EC2), but I disabled it again since it uses like 70MB of our already-limited memory.
• Adds documentation for deployments to AWS via Terraform.