Add blocked scalar spmm kernels.

Remove unrolled scalar kernels as the pipeline version is superior.

PiperOrigin-RevId: 276587361
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 0df9150..214a3ed 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -198,17 +198,15 @@
   src/f32-prelu/x4-scalar.c
   src/f32-rmax/scalar.c
   src/f32-spmm/1x1-scalar-pipelined.c
-  src/f32-spmm/1x1-scalar-unroll2.c
   src/f32-spmm/1x1-scalar.c
   src/f32-spmm/2x1-scalar-pipelined.c
-  src/f32-spmm/2x1-scalar-unroll2.c
   src/f32-spmm/2x1-scalar.c
   src/f32-spmm/4x1-scalar-pipelined.c
-  src/f32-spmm/4x1-scalar-unroll2.c
   src/f32-spmm/4x1-scalar.c
   src/f32-spmm/8x1-scalar-pipelined.c
-  src/f32-spmm/8x1-scalar-unroll2.c
   src/f32-spmm/8x1-scalar.c
+  src/f32-spmm/8x2-scalar.c
+  src/f32-spmm/8x4-scalar.c
   src/f32-vadd/scalar.c
   src/f32-vmul/scalar.c
   src/f32-vmulcaddc/c1-scalar-x2.c