Optimize nested populate

In the following `PostSchema` we have nested structure that has `_author` field in post itself and comments:

```
var PostSchema = mongoose.Schema({
_author: {
type: mongoose.Schema.Types.ObjectId,
ref: 'User',
},
comments: {
type: [{
_author: {
type: mongoose.Schema.Types.ObjectId,
ref: 'User'
},

text: {
type: String
}
}]
}
});

var PostModel = mongoose.model('Post', PostSchema);
```

When we fetch post and populate `_author` fields mongoose makes two queries:

```
PostModel.findById(ObjectId(...)).populate('_author comments._author').exec(callback);

Mongoose: users.find({ _id: { '$in': [ ObjectId("1") ] } }) { fields: undefined }
Mongoose: users.find({ _id: { '$in': [ ObjectId("2") ] } }) { fields: undefined }
```

It would be great performance increment if mongoose will make only one query to populate `_author` field:

```
Mongoose: users.find({ _id: { '$in': [ ObjectId("1"), ObjectId("2")] } }) { fields: undefined }
```

No source files or tests are named. Start by tracing how populate handles the top-level _author and comments._author paths, then verify the query logging for the issue's PostSchema example. Done means both references to User are batched into one users.find query without changing populated results.

Written by the indexing model from the issue text.