Adjust the max iteration count wrt arithmetic precision

Submitted by Gael Guennebaud @ggael

Assigned to Nobody

Link to original bugzilla bug (#479)

Description

Currently, the maximal number of iterations allowed in the Schur decomposition is fixed to 40. While this is good for double, this is clearly not enough for floating point numbers with much higher precision. The same situation probably arises in other iterative algorithms.

I see several options:

1 - find a way to automatically compute a max number of iterations based on NumTraits<Scalar>::epsilon()

2 - let the user adjust the max number of iterations

3 - find another way to detect no convergence

Blocking

#387 (closed)