Feature: compute module_gint by GPU

[dzzz2001]

Linked Issue

fix #3809, fix #3424, fix #3426, fix #3430

Unit Tests and/or Case Tests for my changes

• Unit tests have been added for "spherical harmonics" and "vbatch_matrix_mul".

What's changed?

• Add CUDA code for grid integration, now we can compute grid integrals with gamma_only=1 using GPU.
• We have also maintained the cusolver related code, now we can use cusolver to solve eigenvalues.

Any changes of core modules? (ignore if not applicable)

• Add CUDA code to module_gint.
• Modify the cusolver related code in module_hsolver.

[caic99]

Hi @dzzz2001 , To close multiple issues, prepend "fix" to each of them.

[haozhihan]

Hi @dzzz2001, I noticed that the content regarding cusolver has not been modified in the docs (ks_solver in the input-main.md file)

[haozhihan]

And maybe, you should rewrite the description about abacus gpu version in docs.

[caic99]

Building will take nearly half an hour, and I wonder if code_gen.cpp is required. @dzzz2001