CI: Unpin NumPy in the GMT Dev Tests workflow and fix the dataarray_to_matrix doctest (#3208)

[seisman]

• Unpin NumPy so that NumPy 2.0.0 is installed
• Let pygmt.clib.conversion.dataarray_to_matrix return a list of float objects (Closes Failing doctest on pygmt.clib.conversion.dataarray_to_matrix with NumPy 2.0 #2628)

[seisman]

The "GMT Dev Tests" workflow passed four days ago (https://github.com/GenericMappingTools/pygmt/actions/runs/9508848053/job/26210865246) but suddenly failed 12 hours ago (https://github.com/GenericMappingTools/pygmt/actions/runs/9540235948/job/26291686980).

There are no changed on the GMT side, so it's likely due to a broken library installed by conda-forge.

I think we can go with this PR and find out why the workflow fails on Windows later.

[weiji14]

Thanks @seisman, I think we can go with this to handle compatibility between NumPy 1.x and 2.0.

Btw, do you want to bump the NumPy version on ci_tests.yml after this? Or we keep testing NumPy 1.26 for a few more months?

pygmt/.github/workflows/ci_tests.yaml

Line 78 in 3719474

numpy-version: '1.26'

[seisman]

Btw, do you want to bump the NumPy version on ci_tests.yml after this? Or we keep testing NumPy 1.26 for a few more months?

Tried in 3fc0156, but there are dependency issues
https://github.com/GenericMappingTools/pygmt/actions/runs/9537393685/job/26285444415?pr=3226. It seems some packages are not updated to work with NumPy 2.0

[seisman]

Btw, do you want to bump the NumPy version on ci_tests.yml after this? Or we keep testing NumPy 1.26 for a few more months?

Tried in 3fc0156, but there are dependency issues https://github.com/GenericMappingTools/pygmt/actions/runs/9537393685/job/26285444415?pr=3226. It seems some packages are not updated to work with NumPy 2.0

I think the dependency issue has been resolved. See #3292.