Discrepancies with other test suites

In #36 it was discovered that these implementations differ from CUTEst. In several cases that were checked by hand against the original paper, there were some errors in this package but also at least as many in CUTEst. Consequently one should be agnostic about how one interprets these. For the record (at the time of testing, Oct 12 2019, with 168 problems), here are the discrepancies:
- [X] broydn7d (AMPL model at https://github.com/mpf/Optimization-Test-Problems is wrong)
- [X] brybnd
- [ ] fletchcr (TH: eyeball-check we agree with AMPL, couldn't access paper)
- [ ] genrose (TH: differs from CUTEst only by a constant offset of 1.0)
- [ ] hs46 (TH: eyeball-check we agree with AMPL)
- [ ] hs47 (TH: eyeball-check we agree with AMPL)
- [ ] hs55 (TH: eyeball-check we agree with AMPL)
- [ ] hs70 (CUTEst gives NaN here)
- [ ] hs83 (TH: eyeball-check we agree with AMPL)
- [ ] hs93
- [ ] hs100
- [ ] hs104
- [ ] hs105
- [ ] hs113
- [ ] hs114
- [x] morebv (we were buggy but the remaining discrepancy seems to be a CUTEst bug, see https://www.osti.gov/servlets/purl/6650344)
- [X] nondia (was buggy here too but was fixed in #36; remaining discrepancy is an error in CUTEst and https://github.com/mpf/Optimization-Test-Problems)
- [X] penalty3 (TH: eyeball-check ours looks OK vs http://www.cs.cas.cz/matonoha/download/V1081.pdf)
- [X] powellsg (ours was buggy)
- [X] sbrybnd (there are claims of a typo in Luskan that are fixed here, that agrees with their squaring of the unscaled problem `brybnd`)
- [X] scosine
- [X] sinquad (ours had an error in component 1, fixed, remaining discrepancy is eyeball-correct)

A check mark means that someone has looked at the original paper and decided that the implementation here is correct, or that any discrepancies have been resolved. (Just checking against AMPL does not mean the paper has been checked.)