…th memset since aggregate stores are bad (#55879)

This fixes the issues found in slack in the reinterprets of
```julia
julia> split128_v2(x::UInt128) = (first(reinterpret(NTuple{2, UInt}, x)), last(reinterpret(NTuple{2, UInt}, x)))
split128_v2 (generic function with 1 method)

julia> split128(x::UInt128) = reinterpret(NTuple{2, UInt}, x)
split128 (generic function with 1 method)

@code_native  split128(UInt128(5))
	push	rbp
	mov	rbp, rsp
	mov	rax, rdi
	mov	qword ptr [rdi + 8], rdx
	mov	qword ptr [rdi], rsi
	pop	rbp
	ret

@code_native  split128_v2(UInt128(5))
	push	rbp
	mov	rbp, rsp
	mov	rax, rdi
	mov	qword ptr [rdi], rsi
	mov	qword ptr [rdi + 8], rdx
	pop	rbp
	ret
```

vs on master where
```julia
julia> @code_native  split128(UInt128(5))

	push	rbp
	mov	rbp, rsp
	mov	eax, esi
	shr	eax, 8
	mov	ecx, esi
	shr	ecx, 16
	mov	r8, rsi
	mov	r9, rsi
	vmovd	xmm0, esi
	vpinsrb	xmm0, xmm0, eax, 1
	mov	rax, rsi
	vpinsrb	xmm0, xmm0, ecx, 2
	mov	rcx, rsi
	shr	esi, 24
	vpinsrb	xmm0, xmm0, esi, 3
	shr	r8, 32
	vpinsrb	xmm0, xmm0, r8d, 4
	shr	r9, 40
	vpinsrb	xmm0, xmm0, r9d, 5
	shr	rax, 48
	vpinsrb	xmm0, xmm0, eax, 6
	shr	rcx, 56
	vpinsrb	xmm0, xmm0, ecx, 7
	vpinsrb	xmm0, xmm0, edx, 8
	mov	eax, edx
	shr	eax, 8
	vpinsrb	xmm0, xmm0, eax, 9
	mov	eax, edx
	shr	eax, 16
	vpinsrb	xmm0, xmm0, eax, 10
	mov	eax, edx
	shr	eax, 24
	vpinsrb	xmm0, xmm0, eax, 11
	mov	rax, rdx
	shr	rax, 32
	vpinsrb	xmm0, xmm0, eax, 12
	mov	rax, rdx
	shr	rax, 40
	vpinsrb	xmm0, xmm0, eax, 13
	mov	rax, rdx
	shr	rax, 48
	vpinsrb	xmm0, xmm0, eax, 14
	mov	rax, rdi
	shr	rdx, 56
	vpinsrb	xmm0, xmm0, edx, 15
	vmovdqu	xmmword ptr [rdi], xmm0
	pop	rbp
	ret
```