Optimize Tanner QSPA using runtime metrics

[Parvfect]

Currently is slower than Matrix QSPA although it performs slightly better (hahaha)

But yeah, optimize to get it as quick as you can. It's gonna be decently slow since there is a n^3 term in there but we shall have a look

[Parvfect]

Improved a lot by changing where sum was taken for the normalize function that reduced it to a point where I can run it for long codes for decent iterations. Might be some scope here and there for more improvements.