bitWidth is wrong for -(power-of-two)

```
XCTAssertEqual(64, BigInt(Int64.min).bitWidth)
```

That fails, because bitWidth returns 65.  It's not correctly handling the cases of 0b1 followed by some number of 0b0's, because with two's complement the magnitude of your minimum value is _one higher than of your maximum value_.  Doing `magnitude.bitWidth + 1` fails to account for this.