Why does the number of ``ctlike`` iterations depend on the number of threads used? about gamma-speed

[cdeil]

@jknodlseder Andrei found that the number of likelihood fit iterations depends on the number of OpenMP threads used (in a reproducible way). Do you have an idea why this might be the case?

Since the OpenMP thread parallelism is in the loop over observations, which is inside the likelihood function, I would have expected to always have the same number of likelihood fit iterations, irrespective of the number of OpenMP threads!?

[ignatndr]

I forgot to mention how to reproduce this exact situation:

1. Simulate 3 observations with ctobssim with a duration of 1800 s using the default values for the other input parameters;
2. Make .xml file that repeats these observations a 100 times;
3. Run ctlike on the .xml file.