Improve image sampling implementation

[nyorain]

This implements a new API for image sampling with the following goals:

• Handle sampler details in spvm, expose an interface for specifying them. The sampler definition added here is kept very close to Vulkan, mapping OpenGL onto that shouldn't be too hard. The implementation of the sampling process is mostly a direct implementation of the Vulkan spec (should be more or less similar in OpenGL though) except that this PR doesn't handle some advanced cases (anisotropy, corner-sampled images, multisampling, unnormalized coords).
• Allow applications to handle image reading/writing in a custom manner. Not all applications might have the image as raw data, as it was previously required. It's handled here simply via callbacks in spvm_state.
  • I added an implementation of this API that is used by default and mirrors the old approach.
  • Note that spvm_image is now the "base" struct only containing basic image information and applications will extend that struct with their own information. The implementation mirroring the old approach is called spvm_image_data so we have an API incompatibility here. If that's a big issue we could change the names (using something like spvm_image_base for the interface instead) but I'd rather not.

More of a RFC on the API design for now, the implementation still has some TODOs, isn't well tested and I might have missed one or two changes from my modified spvm version that are already needed. I would be in favor of adding a small set of unit tests to spvm and be willing to write them.

This PR already contains some additional fixes that were needed/useful to implement the new sampling:

• We now always store the correct typeid in values (wasn't the case e.g. for vector values before)
• It seems that spvm previously allocated additional members for scalar values that are part of a struct. Those were not needed and never accessed and are removed now.
• added a SPVM_ prefix to the macros. Surprisingly many projects think it's a good idea to define macros like MIN in their header files.
• seperate between image, sampler and sampled_image values. A sampled_image consists of two members: first the image, then the sampler.
• add an assert in spvm_member_memcpy that was useful during debugging and helps code documentation.
• my editor once again removed some whitespace before eol. Could split it off into a separate commit if it's an issue.

[dfranx]

Thank you so much for the PR. I've tested it a bit and it works nicely. I am going to give it a more thorough test once I integrate this new version of SPIRV-VM into SHADERed.

I am open to unit tests too -- it would certainly improve the quality of the project!

Anyway, it's a really nice PR, thank you once again. I've added you to the README.md as a contributor.