Massive CAL_nsamp slows down entire projection

[quang-huynh]

I have an OM with massive CAL_nsamp = 25000 which I had believed was purely for cosmetic reasons, but it turns out to massively slow down the projections (100 MOM simulations) because MSEtool:::getfifth will generate a 25,000 length vector to calculate the 5 percentile of the length comp.

The easiest solution is to avoid giant nsamp in the first place, but alternative solutions for getfifth could be sought to get around it. The density function appears to be most robust in the edge cases (compared to approx) but is also dependent on sample size

[quang-huynh]

Robust code for getfifth is moderately urgent because AddRealData updates CAL_nsamp. Obviously, someone can condition an OM by passing in a length comp matrix with massive numbers