Using `saturating_sub()` before converting to `usize` gives a wrong
feeling of security as it looks like it ensures that the value will
never go negative. However, since it is applied to `i32`, it can, and
converting it to `usize` would go horribly wrong anyway.

By following the code flow, `exponent` cannot be greater than
`precision`, or the `else` block would not have been taken.  A plain
subtraction will give the same result and will at least panic in debug
mode.

`seq --format %.2g 10 10` would display `1` because the precision would
not allow room for the decimal point, and the `0` in `10` would be
trimmed as an insignificant trailing `0`.

This has been fixed by only trimming trailing `0` in the presence of a
decimal point.