Instead of using a batch size of 1, divide the batch size by 3 so that
we have enough room for 3x more primitives to render. Also use local
Triangle data structures to copy the vertices for the extra primitives,
instead of adding them to the input batch.

Bug: b/139872671
Change-Id: I1bc860d291b53fdd266b6c018ac0c47c876aaa09
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/35588Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>