Capistrano roles ignored for symfony_command

| Library | Version |
|---|---|
| Capistrano | 3.12.1 |
| Capistrano_symfony | 2.0.0 |

I followed this piece from the docs:

```ruby
If you want to execute a command on a host with a given role you can use the Capistrano on DSL, additionally using within from Capistrano will change the directory

namespace :deploy do
task :migrate do
on roles(:db) do
symfony_console('doctrine:migrations:migrate', '--no-interaction')
end
end
end
```

I have 2 servers set up:
```ruby
server "xxx",
user: "emc",
roles: %w{web db},
primary: true,
ssh_options: {
user: "www-data", # overrides user setting above
keys: %w(config/azure_ssh),
forward_agent: false,
auth_methods: %w(publickey)
}

server "xxx2",
user: "emc",
roles: %w{web},
ssh_options: {
user: "www-data", # overrides user setting above
keys: %w(config/azure_ssh),
forward_agent: false,
auth_methods: %w(publickey),
port: 23
}
```

Yet the `symfony_console` is still ran on both servers. I noticed another task which did neatly run on just one server, it was not using the symfony command. So I changed my symfony_command to a capistrano `execute` and now the role is actually taken into account.

For some reason, when using the symfony_command, the roles from capistrano are ignored...

Start by reproducing the deployment with the two server definitions and compare symfony_console with the Capistrano execute call inside the on roles(:db) block. Trace how symfony_console selects its execution hosts; done means the Symfony command runs only on servers assigned the requested role.

Written by the indexing model from the issue text.