Implement simd_funnel_shl and simd_funnel_shr intrinsics

[0xmuon]

Summary

rustc_codegen_cranelift does not implement the simd_funnel_shl and simd_funnel_shr intrinsics. Code using them fails to compile with:

error: Unknown SIMD intrinsic simd_funnel_shl
error: Unknown SIMD intrinsic simd_funnel_shr

This blocks the rustc test tests/ui/simd/intrinsic/generic-arithmetic-pass.rs, which is currently skipped in scripts/test_rustc_tests.sh with the comment unimplemented simd_funnel_{shl,shr}.

Reproduce

#![feature(repr_simd, core_intrinsics)]
#![allow(non_camel_case_types)]

use std::intrinsics::simd::{simd_funnel_shl, simd_funnel_shr};

#[repr(simd)]
#[derive(Copy, Clone)]
struct i32x4([i32; 4]);

fn main() {
    let x = i32x4([1, 2, 3, 4]);
    let y = i32x4([2, 3, 4, 5]);
    unsafe {
        let _ = simd_funnel_shl(x, y, x);
        let _ = simd_funnel_shr(x, y, x);
    }
}

rustc -Zcodegen-backend=cranelift repro.rs

btw simd_shl and simd_shr are already implemented in src/intrinsics/simd.rs
Funnel shifts combine bits from two vectors according to a per lane shift amount

[Sonike]

I am working on this. I will reproduce the missing-intrinsic failure, add a lane-wise implementation for both funnel shifts, re-enable the upstream generic arithmetic test, and run the focused backend tests before opening a PR.