Render pass dependencies

[crud89]

Describe the pull request

This PR improves handling of render pass dependencies, effectively serving as a bare-bone render graph (albeit without support for non render-target dependencies, e.g. resources coming from compute passes, transfers or ray-tracing pipelines, etc.). To achieve this a bit of refactoring has been done:

• Input attachment mappings have been renamed to RenderPassDependency, which appears more meaningful.
• Dependencies now store a binding point, removing some implicit assumptions about where input attachments are mapped to. Note that it is still not possible to have different binding points for different pipelines in the same render pass (although this appears as a minor limitation).
• In a similar fashion, a binding point for the sampler that samples input attachments can now be specified. This removes the requirement of having to specify a root signature just to support a static sampler whenever render passes should be used. Note that the default sampler is opaque to the application (as Vulkan does not need a sampler anyway), but it is also no longer a static sampler (thus requiring the allocation of one sampler descriptor per pipeline instance).
• Render passes can now be configured with render areas different from the swap chain more easily.
• Render pass dependencies cause resize events to be propagated to render passes automatically, if no explicit render area is specified. This removes the requirement of having to manually resize frame buffers for render passes.
• Render pass dependencies also cause resize event propagation to dependent pipelines, which is used to automatically update input attachment bindings.
• On that same note, render pipelines now manage descriptors for binding input attachments on their own. It is no longer required to manage or bind them manually.