Sets up the docker-compose config and deployment scripts to take advantage
of the new (or impending) docker-compose `--env-file` option. Creating a
.env.ENVIRON file then passing the name of the environment to `scripts/deploy`
should read that environment file into the project environment.
All the variables needed inside the tiler container have to be listed in
it's `environment` so they get passed through.

Adds a checked-in .env.local file, with credentials for the test database.
It doesn't need any of the Lambda or project variables, since it's local.

Changes the deployment-related scripts to expect an argument for which
environment to deploy. They'll look for a `.env.ENVIRON` file to set the
environment variables and will create claudia-ENVIRON.json and
.api-id-ENVIRON files rather than always using the same unqualified names.

Using separate claudia.json files means the config file is now given as
an argument, rather than using the default, and they go in a subdirectory.

Changes the terraform deployment steps to use local tfplan and tfstate
files to keep track of separate environments. The "local" backend is
the default, but specifying it explicitly enables specifying parameters
for it, i.e. the ENVIRON-based filenames to use.

This also switches from using `plan -auto-apply` to doing separate plan
and apply commands, though they're still within the same `publish` script.
The behavior will end up the same, but it now gives you 10 seconds to
bail out if the plan looks weird, rather than applying it as soon as it's
calculated.