Allow customization of record field unpacking for TaggedObject

Currently the only way to get record field unpacking when using a sum-of-records is to avoid intermediate types and use the horrible misfeature that is partial record fields.

```
data Vehicle = Car { make :: String, speed :: Int } | Bicycle { brand :: String, gears :: Int }
deriving (Generic, ToJSON)

encode $ Car { make = "MINI", speed = 150 }
> {"tag":"Car","make":"MINI","speed":150}
```

vs:

```
data Vehicle = VehicleCar Car | VehicleBicycle Bicycle
deriving (Generic, ToJSON)

data Car = Car { make :: String, speed :: Int }
deriving (Generic, ToJSON)

data Bicycle = Bicycle { brand :: String, gears :: Int }
deriving (Generic, ToJSON)

encode . VehicleCar $ Car { make = "MINI", speed = 150 }
> {"tag":"VehicleCar","contents":{"speed":150,"make":"MINI"}}
```

If `TaggedObject` had an `unpack :: Bool` parameter users would be able to decide if they want unpacking or not, using `contentsFieldName` when it's `False`. Alternatively you could make `contentsFieldName` a `Maybe String`, this would just prevent any kind of error recovery if one of the sub-objects serializes to something other than a record, although it's unclear if such error recovery would be desirable anyway.

Start by reading the TaggedObject implementation and its contentsFieldName handling. Compare the proposed unpack :: Bool and Maybe String APIs, then verify the chosen behavior for record and non-record sub-objects; done means users can select unpacked or contents-field encoding without breaking existing behavior.

Written by the indexing model from the issue text.