[midend] Support scalable vector for matmul-transpos-b vectorization

[JuniMay]

This PR adds scalable vector support to matmul-transpose-b vectorization. vec-size is renamed to vf (as vectorization factor), and scalable option is added to enable scalable vector.

The step is now computed with vscale * vf and a symbol is added into the affine maps.

The performances are just the same compared to static vector method on musepi:

----------------------------------------------------------------------------
Benchmark                                  Time             CPU   Iterations
----------------------------------------------------------------------------
benchMatmulTransposeB/matmul             333 ms          333 ms            2
benchMatmulTransposeB/matmul_opt0        108 ms          108 ms            6
benchMatmulTransposeB/matmul_opt1        109 ms          109 ms            6
------------------------------------------------
MatmulTransposeB-opt0 PASS
MatmulTransposeB-opt1 PASS
------------------------------------------------

where opt0 is static vector with vf=32 and -mllvm --riscv-v-vector-bits-min=256 appended to clang args, and opt1 is scalable with vf=8 (without explicit vlen arguments).

The tail processing is kept the same for now. I will resolve conflicts when it is modified (if there are any).