Wrong CALT outputs for sample size 512 and 1000

[johcarter]

the estimates for sample size 512 and 1000 are wrong (very large ) for piwind @hchagani-oasislmf
to replicate run piwind with 1000 samples and view gul_S1_alct report
please can you take a look when you get back? cheers

Originally posted by @johcarter in #294 (comment)

[hchagani-oasislmf]

I was not able to recreate the very large numbers you saw @johcarter. However, a sample subset size of 512 should not have been possible given a total sample size of 1000, as the samples in each subset are non-overlapping. Therefore, a subset size of 512 would range from sample 512 to (512 + 512 - 1) = 1023. This is 23 samples greater than the total sample size, so accessing these samples as elements in the vector of losses can return junk. In your case, it may have returned very large values, thus skewing the estimates. In my case, it may have returned very low values, thus making little difference.

Regardless, the behaviour is incorrect. I have fixed this, and this should drop the 512 samples subset and give you the correct estimates for 1000 samples.