fix: only bind to Unix socket on Heroku with nginx buildpack

[mroderick]

Problem

config/puma.rb was checking for the presence of config/nginx.conf.erb to decide whether to bind to a Unix socket or a TCP port. Since the nginx config file is committed to the repository, this check evaluated to true in all environments — including development — causing Puma to bind to unix:///tmp/nginx.socket instead of a TCP port.

This meant that running bundle exec rails server in development would listen on a Unix socket that is not accessible via a browser, breaking the local development workflow.

Root cause

The condition was introduced in PR #2629 to support the heroku-community/nginx buildpack, which uses bin/start-nginx to launch nginx alongside Puma. The socket and /tmp/app-initialized signal are only meaningful when that buildpack orchestrates the boot. However, checking File.exist?("config/nginx.conf.erb") as a proxy for "running under nginx" is unreliable because the file exists in development, CI, and everywhere the repo is cloned.

Fix

Add a check for ENV["DYNO"] before the socket binding. Heroku sets this variable on all dynos, making it a reliable signal that the nginx buildpack is actually in use.

# Before
if File.exist?("config/nginx.conf.erb")

# After
if ENV["DYNO"] && File.exist?("config/nginx.conf.erb")

This preserves the nginx buildpack integration on Heroku while restoring the default TCP port binding in development and other non-Heroku environments.

Verification

• bundle exec rails server in development binds to http://localhost:3000
• Heroku deployment still works with the nginx buildpack