Apply specialization to shader reflection

[Rua]

Changelog:

- - Specialization constants are now provided with a `HashMap` containing `SpecializationConstant` enum values. The `SpecializationConstants` trait is removed, and `vulkano_shaders` no longer generates structs for specialization constants.
+ - Specialization constants are now provided by calling `ShaderModule::specialize` with a `HashMap` containing `SpecializationConstant` enum values. This produces a `SpecializedShaderModule` value, which you can then create an `EntryPoint` from.
+ - The `SpecializationConstants` trait is removed, and `vulkano_shaders` no longer generates structs for specialization constants.

### Bugs fixed
- Specialization constants are now applied to the reflected SPIR-V code before any other reflection is performed.

Shader specialization has been moved to its own type, SpecializedShaderModule. You create this from a regular ShaderModule by calling with_specialization, and it has methods to create an entry point. The old entry point methods that were directly on ShaderModule are still there, but they are now simply shortcuts for calling with_specialization with no specialization constants, and then calling the corresponding entry point method of SpecializedShaderModule.

When you create a SpecializedShaderModule, a replacement pass is made through the reflected SPIR-V module, replacing all specialization constants with regular constants, applying the provided specialization constants where applicable. OpSpecConstantOp instructions are also evaluated in this pass, which was by far the most complex part of this PR, and also the most likely to have bugs I missed.

[marc0246]

This doesn't fix #1956. The issue also isn't a bug, so the label is incorrect. As I commented, the issue is that they are trying to use vulkano-shaders, which generates Rust types at compile-time, with an array that is runtime-sized, which is impossible. From what I can tell vulkano-shaders is working exactly as intended there.

[Rua]

Good point, I had missed that the issue was with Vulkano-shaders specifically. Should we just close it as wontfix then?

[marc0246]

Question is whether they would find vulkano-shaders that doesn't generate Rust types useful.

[marc0246]

Looks good, thanks!