Previously, vertices would be processed in consecutive groups of four
(for SSE/NEON). Now four indices are read from the index buffer.
Reading the input was already a gather operation, but with constant
stride. The vertex cache now performs a scatter. The vertices are
written in reverse order so that the first vertex in a group is always
present in the cache.

Also use 2^32-1 as invalid vertex cache index (corresponds with the
primitive restart index) instead of 0x80000000, since
maxDrawIndexedIndexValue is UINT32_MAX.

Bug: b/27351835
Test: dEQP-VK.glsl.loops.special.do_while_dynamic_iterations.dowhile_trap_vertex
Change-Id: Ic69dbf53c67cbda50e44913ccae91aaca2b86e21
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/32609
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>