Use tree structure to take the minimum of timestep requests

[sgherbst]

At the moment, the minimum timestep is computed as min(min(min(a, b), c), d) rather than min(min(a, b), min(c, d)). We thought that the synthesis tool would automatically optimize this, but that does not appear to be the case. As a result, the delay added by the time manager grows linearly with the number of timestep requests, rather than with the log of that number.