Highly performant inverse square root operations

[LukePearson1]

The inverse square root operations is fundamental to the Ristretto operations. This is due to division. As division and exponentiation are expensive for a CPU to perform, inverse modular operations are used when any such operations need to be programmed.

For example taking 1/(√x), there is both a division and an exponentiation. For this, a modular inverse square root computes 1 multiplied by inverse mod of square root x.

---

There are multiple ways to compute the aforementioned calculation. This issue is opened to explore the various ways we can implement inverse square root. In general, the non-Euclidian methods provide shorter operation times but it doesn't necessarily mean this will be the case for us. This is a result of which sign choices we assign as well as which canonical endings we need to omit.

---

The issue will document which properties we seek from computed as well as benchmarks of various implemented inverse square root algorithms.