fix: inconsistency in the sqrt function

[Amxx]

Fixes #97.

The sqrt function using newton methods with a first guess that is the most significant bit of the sqrt or x. This is done using an algorithm that is similar to mostSignificantBit, but that increase the result by the half of what mostSignificantBit does.

While mostSignificantBit is correct, sqrt uses 0x8 instead of 0x4 for 2**2

---

Note: a cleaner, but slightly more expensive, version of the code could do

function sqrt(uint256 x) internal pure returns (uint256 result) {
        if (x == 0) {
            return 0;
        }

        // Set the initial guess to the least power of two that is greater than or equal to sqrt(x).
        result = 1 << (mostSignificantBit(x) >> 1); 
        
        // The operations can never overflow because the result is max 2^127 when it enters this block.
        unchecked {
            result = (result + x / result) >> 1;
            result = (result + x / result) >> 1;
            result = (result + x / result) >> 1;
            result = (result + x / result) >> 1;
            result = (result + x / result) >> 1;
            result = (result + x / result) >> 1;
            result = (result + x / result) >> 1; // Seven iterations should be enough
            uint256 roundedDownResult = x / result;
            return result >= roundedDownResult ? roundedDownResult : result;
        }
    }

[PaulRBerg]

Hello @Amxx, I'm sorry for the delay.

This is great, thanks so much for spotting this error and opening a PR to fix.