Backend-specific configuration

[kvark]

We need to establish a way for applications to fine-tune the backends up to their needs. In Metal, I can at least think of the following parameters:

• maximum number of active command buffers
• support for immediately recorded command buffers
• auto-enqueuing of command buffers

These are not required for anybody to touch, but resource demanding applications could benefit for a few knobs to play with in order to assist the performance. E.g. while in general immediate recording is great, with Dota it is actually slower due to rare submissions... see #2232 for more info.

On clear thing we could establish here is some sort of configuration API, maybe even a public struct field, on the Instance or PhysicalDevice (specific to backends). Another possible route is via environment variables, e.g. GFX_METAL_IMMEDIATE_RECORDING=0 make quad. I believe the latter is only needed for applications going through gfx-portability, so we might as well just put the envar handling in there.

[grovesNL]

I suppose another possibility is a gfx extension to Vulkan, so we don't have to commit to any particular configuration API/environment variables on the HAL side (at least depending on what happens with #2206)

[kvark]

@grovesNL indeed, but this would be quite painful:

• extensions need to be published and set in stone, while our flags may be more fluent, and the target user (who is shipping an application with gfx-rs always knows the exact set of flags).
• our flags are vastly different between backends, so we'd have to have an extension per backend