Add overflow check

You clam your library to be "safe casting from one type to another in Go", but in the code there are no checks for overflow when casting integer types:

switch s := i.(type) {
	case int:
		return int16(s), nil  // <-- this may overflow
	case int64:
		return int16(s), nil  // <-- this may overflow
	case int32:
		return int16(s), nil
	case int16:
		return s, nil
	case int8:
		return int16(s), nil
	case uint:
		return int16(s), nil   // <-- this may overflow
	case uint64:
		return int16(s), nil   // <-- this may overflow
	case uint32:
		return int16(s), nil   // <-- this may overflow
	case uint16:
		return int16(s), nil   // <-- this may overflow
	case uint8:
		return int16(s), nil
	case float64:
		return int16(s), nil   // <-- this may overflow
	case float32:
		return int16(s), nil  // <-- this may overflow

Adding bounds check like this: https://play.golang.org/p/Jgu2NYg1qi3 would solve the issue.

Start by locating the type-switch entry point that performs the integer and floating-point casts, then compare the existing conversions with the bounds-check example in the issue. Add overflow handling for the affected casts and verify that values within range still convert successfully while out-of-range values no longer silently wrap.