Relative tolerance for algebraic loops

[rruusu]

Description

Algebraic loops are currently only solved using an absolute tolerance that is also used for the integrator.

If the numerical values of the connected variables grow large enough, the absolute tolerance may become unachievable due to floating point precision and associated rounding errors, resulting in excessive effort and warning messages for non-convergence.

It would be useful to apply the relative tolerance values for algebraic loops.

Related issues

Related to #1327.

[rruusu]

I will push some solutions for this to my fork shortly.