gallivm: dEQP-GLES31.functional.ssbo.layout.random.all_shared_buffer.36 timeouts
This test emits a colossal number of store_ssbos with constant offsets (and data), each of which gets unrolled to a loop over invocations. Except that we have only a 1x1x1 grid, and no exec masking during that stage of the shader, so we could avoid bothering llvm with the loop, and massively reduce the runtime. See 591899ee ("gallivm/nir: Add a short circuit uniform-offset mode for load_ssbo/load_shared.") for prior art.