Fix benchmark

[arrjon]

This pull request updates the Lotka-Volterra and SIR benchmark simulators to be in line with the benchmark of Lueckmann. The changes allow for more accurate reproduction of original benchmark tasks:

• Added 'original' option in the subsample parameter for both lotka_volterra.py and sir.py simulators, enabling users to reproduce the exact subsampling easily.
• Changed the construction of time vectors in both simulators by switching from np.linspace to np.arange to get the same time steps as in Lueckmann et. al.
• In lotka_volterra.py simulations are now clipped to a minimum of 1e-10 and a maximum of 10000.0 before applying noise, instead of the log1p, which migth be nicer but deviates from the original benchmark.