Add a fast_matrix_mul_4x4 function with SIMD optimization for the LoongArch64.

[KatyushaScarlet]

1. Add the function fast_matrix_mul_4x4_lsx for LoongArch64.
2. Add the CFLAGS -mlsx and -mlasx for LoongArch64, which enable gcc to build with LSX/LASX (128/256bit SIMD extension for loongson CPU) instructions.

Here is the Unofficial LoongArch Intrinsics Guide: https://jia.je/unofficial-loongarch-intrinsics-guide/migrating_sse/

[hrydgard]

Ideally CrossSIMD.h needs a Loongarch impl too, in the future some more of the SIMD code of the app will be migrated to use that.

[hrydgard]

Nowadays it's more standard to use memcpy for this, compilers optimize it down properly. This usage of unions is UB. However, in practice it does work, so I'll allow it.