OneofValidator should allow oneof fields to be empty

[yuri-sergiichuk]

The current implementation of the OneofValidator does not allow any oneof field to be not set.

The expected behavior is that it is possible to have unset oneof fields, e.g. we have an origin_id fields in the EventContext message that may be empty by its contract:

// Meta-information for an event.
message EventContext {
    ...
    oneof origin {
        // The context of the command which generated this event.
        CommandContext command_context = 2 [(required) = true];

        // The context of the event which generated this event.
        EventContext event_context = 6 [(required) = true];

        // The event was imported using this `ActorContext`.
        ActorContext import_context = 11 [(required) = true];
    }
    ...
    // An ID of the origin messsage.
    //
    // The case of this `oneof` is always the same as the case of `origin` (except for
    // the `import_context`, which does not have a corresponding field in `origin_id`).
    //
    // The `origin` and `origin_id` groups are not merged for the reasons of binary compatibility.
    //
    oneof origin_id {

        // The ID of the origin command.
        CommandId command_id = 12;

        // The ID of the origin event.
        EventId event_id = 13;
    }
    ...
}

[yuri-sergiichuk]

As discussed vocally, the only way to make oneof field required should be to use required_field message option as follows:

message EventContext {
    option (required_field) = "command_context | event_context | import_context";
    ...
    oneof origin {
        // The context of the command which generated this event.
        CommandContext command_context = 2;

        // The context of the event which generated this event.
        EventContext event_context = 6;

        // The event was imported using this `ActorContext`.
        ActorContext import_context = 11;
    }
    ...
}

The required option at the message field level for oneof fields should be treated now as a configuration error.

Also, by default, oneof fields should not be required

[alexander-yevsyukov]

@yuri-sergiichuk please see if #521 addresses this issue. If so, please close.

[yuri-sergiichuk]

Superseded by #521 and fixed in #523 (the new syntax described in the PR).