Optimize swizzle_dyn for LoongArch64 with N is 16 or 32

[heiher]

We can optimize swizzles for 16-byte and 32-byte vectors using LoongArch LSX/LASX shuffle instructions.

For reference, see the Unofficial LoongArch Intrinsics Guide:
https://jia.je/unofficial-loongarch-intrinsics-guide

I've run all swizzle_dyn::lanes_* tests locally, and they all passed.

[programmerjake]

I've run all swizzle_dyn::lanes_* tests locally, and they all passed.

do you have LASX? afaik it isn't tested unless you enable it as a target feature when building portable-simd, since afaict CI doesn't enable it.

[heiher]

do you have LASX? afaik it isn't tested unless you enable it as a target feature when building portable-simd, since afaict CI doesn't enable it.

Yeah. I tested it locally with:

RUSTFLAGS="-C target-feature=+lasx" cargo test -r

000000000005ae80 <_RNvNtNtCs1775If9jsmw_9core_simd9core_simd11swizzle_dyn19loong64_lsx_swizzle>:
   5ae80:	2c0000a0 	vld         	$vr0, $a1, 0
   5ae84:	2c0000c1 	vld         	$vr1, $a2, 0
   5ae88:	0d508000 	vshuf.b     	$vr0, $vr0, $vr0, $vr1
   5ae8c:	72843c21 	vslei.bu    	$vr1, $vr1, 0xf
   5ae90:	71260400 	vand.v      	$vr0, $vr0, $vr1
   5ae94:	2c400080 	vst         	$vr0, $a0, 0
   5ae98:	4c000020 	ret

000000000005aea0 <_RNvNtNtCs1775If9jsmw_9core_simd9core_simd11swizzle_dyn20loong64_lasx_swizzle>:
   5aea0:	2c8000a0 	xvld        	$xr0, $a1, 0
   5aea4:	2c8000c1 	xvld        	$xr1, $a2, 0
   5aea8:	77d40002 	xvori.b     	$xr2, $xr0, 0x0
   5aeac:	77ec0002 	xvpermi.q   	$xr2, $xr0, 0x0
   5aeb0:	77ec4400 	xvpermi.q   	$xr0, $xr0, 0x11
   5aeb4:	0d608800 	xvshuf.b    	$xr0, $xr0, $xr2, $xr1
   5aeb8:	76847c21 	xvslei.bu   	$xr1, $xr1, 0x1f
   5aebc:	75260400 	xvand.v     	$xr0, $xr0, $xr1
   5aec0:	2cc00080 	xvst        	$xr0, $a0, 0
   5aec4:	4c000020 	ret