Fixing miscellaneous issues in gmtregress

[PaulWessel]

Description of proposed changes

Some issues have come to light via this forum post that required some changes:

1. Except for regular regression with misfit measured vertically (-Ey), the Pearsonian coefficient of determination R is not defined (perhaps it is, but not by the same equation; my Google search could not find an equivalent expression for RMA and Ortho regressions - let me know if you can). The output for the userʻs case gave wild R. I now only output R for -Ey.
2. The RLS algorithm of dividing the search angle range by 2 etc. could stop the search too early if the second iteration visited the same (crude) angle step and that happened to be the smallest misfit yet. Then there would be no drop in best misfit and iterations would stop. I know ensure that this cannot happen until the angle step in the search is at least < 0.05 degrees.
3. The latter change refined the fits in minor ways for three examples and hence their PS files have been updated.

[PaulWessel]

Actually, will let r be printed since it is entirely data driven. But R is wrong (going negative).