Improve iterative initialisation procedure

[kyllingstad]

Currently, we perform N zero-length time steps at the beginning of each simulation, where N is the number of slaves, in order to propagate initial values between slaves. This is OK for FMI 1.0, but not for FMI 2.0, where a step size of 0 is explicitly forbidden.

For 2.0, we should instead make use of the new "initialisation mode" and repeated fmi2GetXXX/fmi2SetXXX calls. It should be possible to handle both cases nicely by modifying the SlaveInstance interface a bit (i.e., include the initialisation phase), and only insert zero-length steps as necessary for FMI 1.0 FMUs.