Better methods for simulation

[JoseKling]

Working on #88 I noticed it might have opened the possibility to make the simulation methods more general and maybe simpler.
The mark distributions have their own sample_mark(md::MarkDistribution, t, h::History) method. The idea is to have a sample_time(pp::PointProcess, h::History) method as well (method names to be defined).
sample_time would return the next event after history h or, if the next event time would fall beyond h.tmax, return nothing. With this, we could define

function simulate(pp::PointProcess, tmin, tmax)
  h = History(tmin, tmax) # Empty history
  t = sample_time(pp, h)
  while t
    m = sample_mark(pp.mark_dist, t, h) # Should probably be implemented for point processes, like `sample_mark(pp, t, h)`
    push!(h, t, m)
    t = sample_time(pp, h)
  end
  return h
end

The advantage being that this works for any point process and any mark distribution, no matter if the mark distribution and event times are dependent.
Additionally, sample_time could have a fallback, which would be basically the algorithm from simulate_ogata, but instead of simulating the whole process, it just simulates the next event.
Another possibility would be to have a method inverse_integrated_ground_intensity (or something similar, but not that long) to implement the fallback for sample_time. This is probably more efficient than sampling multiple times until one can be kept, and since we need an additional method anyway (ground_intensity_bound or inverse_integrated_ground_intensity), might as well use this.