Fix zbb/zbkb detection

[dtcxzyw]

This patch uses sext.b and pack to distinguish zbkb from zbb.

// llvm-mc -filetype=obj -triple=riscv64 -mattr=+zbb,+zbkb < tmp.S | llvm-objdump --mattr=+zbb,+zbkb -d -r -
add a0,a0,a0
sext.b a0,a0 # zbb
pack a0,a0,a0 # zbkb

<stdin>:        file format elf64-littleriscv

Disassembly of section .text:

0000000000000000 <.text>:
       0: 00a50533      add     a0, a0, a0
       4: 60451513      sext.b  a0, a0
       8: 08a54533      pack    a0, a0, a0

Tested with qemu:

riscv64-linux-gnu-gcc main.c
qemu-riscv64 -L /usr/riscv64-linux-gnu/ -cpu rv64,zbb=false,zbkb=false ./a.out
qemu-riscv64 -L /usr/riscv64-linux-gnu/ -cpu rv64,zbb=true,zbkb=false ./a.out
qemu-riscv64 -L /usr/riscv64-linux-gnu/ -cpu rv64,zbb=false,zbkb=true ./a.out
qemu-riscv64 -L /usr/riscv64-linux-gnu/ -cpu rv64,zbb=true,zbkb=true ./a.out

It addresses the issue that ruapu incorrectly reported that zbkb is available on Spacemit-K1.
See also llvm/llvm-project#94564.

[dtcxzyw]

@cyyself @sunshaoce Could you guys test this patch on your K1 boards? Thank you!

[zqb-all]

I tested it on K1 and now it can correctly report zbkb=0

[nihui]

Thanks for your contribution !