Make all Numbers *big.Floats by default.

When handling types that came from JSON or msgpack, anything that's
supposed to be a number we try to coerce into being a *big.Float, which
is the only lossless type we can standardise all Numbers on. This
prevents developers from needing to worry about getting Numbers in
different intermediary types based on what value was provided. Unmarshal
will give developers the type in whatever convenient format they ask
for, but will always throw an error in the face of a loss of specificity
(except for float32/float64, in which case it's impossible to know if
accuracy was lost).

Author: Paddy Carver

tfprotov5/tftypes/raw_value.go

@@ -333,15 +333,40 @@ func unmarshalFloat64(in float64, dst interface{}) error {
 func rawValueFromComplexType(typ Type, val interface{}) (RawValue, error) {
 	if _, ok := typ.(primitive); ok {
 		if typ.Is(Number) {
-			// TODO: put numbers in a consistent format
-			// numbers could be coming to us as like a billion
-			// different types, and we want developers to have
-			// stable types to build against. We should cast or
-			// wrap as necessary here to handle that.
-			return RawValue{
-				Type:  Number,
-				Value: val,
-			}, nil
+			result := RawValue{
+				Type: Number,
+			}
+			switch v := val.(type) {
+			case *big.Float:
+				v.Set(v)
+			case uint64:
+				result.Value = big.NewFloat(float64(v))
+			case uint32:
+				result.Value = big.NewFloat(float64(v))
+			case uint16:
+				result.Value = big.NewFloat(float64(v))
+			case uint8:
+				result.Value = big.NewFloat(float64(v))
+			case uint:
+				result.Value = big.NewFloat(float64(v))
+			case int64:
+				result.Value = big.NewFloat(float64(v))
+			case int32:
+				result.Value = big.NewFloat(float64(v))
+			case int16:
+				result.Value = big.NewFloat(float64(v))
+			case int8:
+				result.Value = big.NewFloat(float64(v))
+			case int:
+				result.Value = big.NewFloat(float64(v))
+			case float64:
+				result.Value = big.NewFloat(v)
+			case float32:
+				result.Value = big.NewFloat(float64(v))
+			default:
+				return result, fmt.Errorf("can't use type %T as a %s", val, Number)
+			}
+			return result, nil
 		}
 		return RawValue{
 			Type:  typ,