Given the findings in #2013, it seems like we should do some investigation to determine if validation is needed to ensure compatibility between a pipeline's render target formats and the blend state used. For example: Is it valid to use a blend state of dst-alpha with a target state format of rg8unorm, since that format ostensibly doesn't have an alpha channel to sample from?

Ideally the underlying implementation would treat the missing alpha channel as 1, but we should verify that this is the case and if cross-platform differences are found some validation should be added to avoid differences in behavior across devices.