WIP: intel: Use OWord block messages for uniform indirect UBO pulls

VERY VERY WIP

Currently, the Intel back-ends bucket UBO pulls into two categories: constant-offset and indirect. For constant-offset UBO pulls we use OWORD block messages which can read up to 128B of data at a time. For indirect UBO access we use untyped surface messages or the sampler. However, there's a third category that we're not currently taking advantage of: indirect UBO pulls where the indirect offset is uniform. In this case, we can still use the OWORD block messages if we're careful and get more data in a single pull. This also reduces register pressure because we aren't splatting out the data across channels.