Using inject breaks enumerations with proto3

[Cohedrin]

When using inject alongside enumerations, the namespace of the enumeration modules is inconsistent, causing the call to get_default to error with a no match, coming from here.

Reference protobuf

syntax = "proto3";

message Message {
  enum Status {
    STATE_1 = 0;
    STATE_2 = 1;
  }

  required int64 id = 1;
  optional Status status = 2;
}

Example

Module

defmodule MessageTest do
  use Protobuf, from: Path.expand("priv/protos/message.proto"), inject: true, only: :Message
end

Defs

iex(1)> MessageTest.defs()
[{{:enum, :"Message.Status"}, [STATE_1: 0, STATE_2: 1]},
 {{:msg, :Message},
  [%Protobuf.Field{fnum: 1, name: :id, occurrence: :required, opts: [], rnum: 2,
    type: :int64},
   %Protobuf.Field{fnum: 2, name: :status, occurrence: :optional, opts: [],
    rnum: 3, type: {:enum, MessageTest.Message.Status}}]}]

Because of mismatch of the naming (MessageTest.Message.Status in the field vs :enum, :"Message.Status" in the key), when the field is looked up in proto3_type_default, lists:keyfind/3 returns false, rather than a match.

I was able to circumvent this by adding use_module_namespace: true, unfortunately this has the side effect of ignoring the inject and only option, as there is no function clause that takes it into account

When not using inject, the field defs are generated as expected

Example

Module

defmodule MessageTest do
  use Protobuf, from: Path.expand("priv/protos/message.proto")
end

Defs

iex(1)> MessageTest.defs()
[{{:enum, MessageTest.Message.Status}, [STATE_1: 0, STATE_2: 1]},
 {{:msg, MessageTest.Message},
  [%Protobuf.Field{fnum: 1, name: :id, occurrence: :required, opts: [], rnum: 2,
    type: :int64},
   %Protobuf.Field{fnum: 2, name: :status, occurrence: :optional, opts: [],
    rnum: 3, type: {:enum, MessageTest.Message.Status}}]}]

I'm happy to make a pr solving this issue, but would like to get some feedback on the direction to take with this before making a pr.

[bitwalker]

If you open a PR with a fix, I'll be happy to work with you on it there, I think we'll have to iterate on it a bit, but if you kick things off I can pick up where you leave off if time is an issue.