New and improved executor

[christophermaier]

Uses the new parser and simplifies A LOT of what the executor does.

Introduces a new concept of an "execution plan" for pipeline stages,
in an attempt to decouple from the parser AST (though that's not quite
complete just yet).

Lots more remains to be done, but this takes care of a huge chunk of
the work.

Closes #286

[vanstee]

No need to wrap prepare_option in an anonymous function. Could just use &prepare_option/2

[vanstee]

Might be a good place to use with

defs = Enum.reduce(command_model.options, %{}, &(prepare_option(&1, &2)))

with {:ok, options, args} <- interpret(defs, args, [], %{}),
     :ok <- check_required_options(defs, options) do
  options = set_defaults(invocation, options)
  {:ok, options, args}
end

Also not sure if defs can ever not be a map so that first case might be unnecessary.

[vanstee]

The only two important comments were the ones about creating a request struct and whether or not we're rendering a template per item in the output list. Other than that this looks great. Seriously good work @christophermaier.

[kevsmith]

💯 @christophermaier !

If we can all agree the new template rendering logic is correct I'm 👍 to merge. Let's table the request/response struct to a new ticket.

[lhaskins]

I'm not following how the anonymous func is getting the namespace and command name from the inputs passed. Maybe more comments about how this works.

[mpeck]

Piper invokes the anonymous function with the namespace and command name @lhaskins. If that helps at all?

[christophermaier]

@lhaskins command_resolver_fn returns a function that closes over the passed in user. This function is passed into the parser and is used as a callback to lookup the relevant information; the parser determines the bundle name and command name to be passed in based on the parsed text.

[lhaskins]

Ah! gotcha - a callback function that gets executed in Piper. Thanks!