Make fitting of the GOK more robust in calc_Huntley2006()

[mcol]

When fitting the simulated curve with the GOK model fails, the fit is attempted again with 100 different values for D0 and the best fit is used. This should reduce the number of occasions in which the error message "Could not fit simulated
model curve, check suitability of model and parameters" is reported.

When I made the simulated curve of the testcase work, I started getting an error in the fitting of the unfaded curve (the fit failed with "Missing value or an infinity produced when evaluating the model" from nlsLM). I reckon that what happens is that we take the starting values from the exponential model, and in that case the c parameter was negative, so at first I took its absolute value, but then I settled on the following change, which fixed that error and got a solution for the testcase:

-   c = coef(fit_start)[["c"]],
+   c = max(coef(fit_start)[["c"]], 1),

Fixes #660.

[RLumSK]

Thanks @mcol. Could you make als dataset I had sent work?

[mcol]

The testcase I've added is derived from the first dataset you sent me, just with values rounded to 2 decimal places. Your original dataset works too, just very slowly (~30 minutes) as it asks for 10000 MC iterations, and each of them now has to fit 100 models for the simulated curve. I got results testing with n.MC = 100 in ~20 seconds.

[RLumSK]

Upon thorough examination, particularly regarding the extended execution times, it becomes evident that we should consider dividing the computation into distinct processes to harness the capabilities of multicore systems.

[mcol]

I've now parallelized the MC computations. The annoying thing is that we cannot have calls to .throw_error() inside the parLapply() block as it generates this sort of messages:

Error in checkForRemoteErrors(val) :
  one node produced an error: attempt to select less than one element in integerOneIndex

The only workable solution I found is to return NA and check outside of the parallel region for errors. Therefore we lose the advantage of returning early in case of lack of fit, as we need to wait for the whole MC run to complete.

[mcol]

This takes the time to run 10,000 MC iterations on your example with 10 cores to ~5 minutes.

[RLumSK]

Great @mcol!