LoongArch: Fix shared cache size calculation

Current calculation of shared cache size is from the node (die) scope,
but we hope 'lscpu' to show the shared cache size of the whole package
for multi-die chips (e.g., Loongson-3C5000L, which contains 4 dies in
one package). So fix it by multiplying nodes_per_package.

Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>