Shader module unchecked doesn't disable bounds checking codegen

[unfolding18]

Description
Naga adds additional bounds checking code to WGSL shaders prior to compilation. It is configurable via Naga, but does not appear to work in wgpu-rs itself. In particular, create_shader_module_unchecked() does not disable bounds checking. Note that I have only checked this on the Metal backend, as that is the hardware I am using/targeting.

Repro steps
Create a WGSL shader that indexes into any array, buffer, matrix, or vec. Set wgpu to use the Metal backend. Run the executable through Xcode's shader debugger. You should see that, regardless of whether you use create_shader_module or create_shader_module_unchecked, the resulting Metal shader code contains various conditionals and bounds checking routines.

Expected vs observed behavior
Ideally, create_shader_module_unchecked would disable bounds checking (or this would be configurable somehow via the API).

Platform
MacOS Ventura 13.2, wgpu version 0.14.0, Metal backend.

[cwfitzgerald]

This appears to be a two sided problem. First is that we just are never changing the bounds check policies we pass to naga. See

wgpu/wgpu-hal/src/metal/device.rs

Lines 692 to 698 in e444520

	bounds_check_policies: naga::proc::BoundsCheckPolicies {
	index: naga::proc::BoundsCheckPolicy::ReadZeroSkipWrite,
	buffer: naga::proc::BoundsCheckPolicy::ReadZeroSkipWrite,
	image: naga::proc::BoundsCheckPolicy::ReadZeroSkipWrite,
	// TODO: support bounds checks on binding arrays
	binding_array: naga::proc::BoundsCheckPolicy::Unchecked,
	},

for where they are hard coded. Second we are setting them in the pipline layout, when the unchecked-ness comes on a per-module basis.