Marshal* should return the tail... #1

Marshal() currently returns the whole bytes buffer which makes it awkward to use correctly.

If you want write a k/v pair you have resort to error-prone hacks like:

val, _ := sb.Marshal(msg)
sp := sb.Len()
key := sb.PutUint64(123).Bytes()[sp:]

A better API is probably something like making all methods return a Result:

type Result struct {
    *Buf
    Err error
    Dat []byte
}

This way chaining works and you can easily extract the data written by the last operation.