Add RBFSinglePrecision for a single precision option

[Icbears]

Introduced RBFSinglePrecision class for memory optimization.
edit_mesh.zip
test.zip

[Icbears]

Introduced RBFSinglePrecision class for memory optimization. edit_mesh.zip test.zip

A fluent mesh sample is attached, along with the code for mesh deformation.

[ndem0]

Dear @Icbears, thank you! The PR is fine, but looking at the code, I've seen that the new class actually introduces a parameter for the precision (float32/64). To avoid code repetition, I ask you to add that parameter in the __init__ in the existing RBF class. It's more maintainable and in principle it allows us also to use other types (complex, float128, ...)

[Icbears]

Dear @ndem0, thank you for your advice. I have added a dtype parameter (default set to fp64) to the RBF class, allowing users to select from the following precisions: fp16, fp32, fp64, and fp128. I tested these data types on my devices, and while fp16 consumes even less memory, it may give inaccurate mesh outputs. Additionally, my Windows 11 laptop does not support np.float128, although it works on a Linux workstation. By the way, the RBFSinglePrecision class was removed because it may be redundant. Please see the changes from the file changed section, and feel free to let me know your thoughts.