riscv: P extension intrinsics for packed SIMD (part 1)

implement by inline assembly for now, uses `pure, nomem, nostack` for
all packed simd arithmetic instructions. Uses `inlateout` when it
requires using the same register for input and output.

Author: luojia65

crates/core_arch/src/riscv_shared/mod.rs

@@ -1,4 +1,7 @@
 //! Shared RISC-V intrinsics
+mod p;
+
+pub use p::*;
 
 use crate::arch::asm;