AVX2: Use mulcache for base multiplication

[hanno-becker]

The AVX2 backend should use a mulcache like the C and AArch64 backends. This essentially means moving

mlkem-native/mlkem/native/x86_64/basemul.S

Lines 90 to 105 in 7294554

	* TODO: This could be precomputed in the mulcache */
	vmovdqa (%r9),%ymm0
	vmovdqa 32(%r9),%ymm1
	vpmullw %ymm0,%ymm10,%ymm2
	vpmullw %ymm0,%ymm12,%ymm3
	vpmulhw %ymm1,%ymm10,%ymm10
	vpmulhw %ymm1,%ymm12,%ymm12
	vpmulhw %ymm8,%ymm2,%ymm2
	vpmulhw %ymm8,%ymm3,%ymm3
	vpsubw %ymm2,%ymm10,%ymm10 # rb0d0
	vpsubw %ymm3,%ymm12,%ymm12 # rb1d1
	vpaddw %ymm5,%ymm9,%ymm9
	vpaddw %ymm7,%ymm11,%ymm11
	vpsubw %ymm13,%ymm10,%ymm13
	vpsubw %ymm12,%ymm6,%ymm6

into a new poly_mulcache_compute_native routine.