Simplify the tests for the greenspline module

Two of the 10 slowest tests are for the greenspline module (gspline_2.sh and gspline_6.sh). However, a small fraction of the time for these two tests is actually spent in the greenspline module (~0.2s greenspline, ~21s surface, ~0.5s plotting + graphicsmagick time). Here are a couple ideas for simplifying these tests to take less time:

1. Use ncdump or a similar tool to compare the output from greenspline to a reference grid file, removing the need for surface or plotting modules completely.
2. Use either DVC or the GMT cache to store the output from `gmt surface @Table_5_11.txt -R$R -I$D -Graws5.grd -T0.5 -N10000 -C0.0000001`, such that the slow surface step does not take place in the greenspline tests.

I would prefer using option 1 for the daily tests of greenspline. Since the tests also serve to guarantee that the figures in Paul's 2009 paper are reproducible, perhaps we could have a separate set of tests for published figures that can be optionally run (similar to the API tests and the proposal for the supplement tests in #6324). Any thoughts @GenericMappingTools/gmt-maintainers?