Provide native-only defaults for Limits

[Bromles]

Is your feature request related to a problem? Please describe.
Currently, we have several options for specifying limits:

• downlevel_defaults, which is guaranteed to work everywhere except on WebGL
• downlevel_webgl2_defaults, which is guaranteed to work even on WebGL
• defaults, which is guaranteed to work on primary backends, including WebGPU

However, wgpu currently has many native-only features, such as push constants, binding arrays, and others. Also, there is a possibility for web-compatible limits to be higher on the native (e.g., a higher max bind group count).

The convenient methods mentioned above set native-only limits to 0, since they are incompatible with WebGPU and downlevel backends.

Describe the solution you'd like
It would be nice to add a native_defaults method to the Limits struct to provide guaranteed values for the native platforms, including native-only features and appropriate native limits

Describe alternatives you've considered
If that's not possible, we could at least specify expected values for each limit field, since some fields - such as min_subgroup_size - do not currently have them

Additional context

[inner-daemons]

Many of the native features aren't supported on every native device. It doesn't make sense to have a set of limits predefined that won't work on many native devices and where most users wouldn't have a need for many of the features.

What would perhaps be more useful is to add more functions like using_recommended_minimum_mesh_shader_values.

[Bromles]

That is also a very valid way to solve the problem.

Honestly, it doesn't really matter how the limits are provided exactly, as an overall method for filling the whole struct, or as a feature-specific methods like the one linked by you

What matters is the existence of a convenient way for the developer to pick reliable values without the need to determine them for every backend themselves

[Bromles]

Also, I believe the docs for the min_subgroup_size and max_subgroup_size are mixed up as well. Elaborated on it in the #8335

Do I need to open a separate issue for this?

[inner-daemons]

Thanks for finding these issues with the limits. If you can, I think it would be beneficial if you could create a PR to address the problems, as I'm not sure it would be anyone else's number 1 priority.