[dev.simd] simd: add partial slice load/store for 32/64-bits on AVX2

[dev.simd] simd: add partial slice load/store for 32/64-bits on AVX2

These all use int-vector-masked loads and stores.
Partial set of tests (for all NxK shapes, thought not all types).

Change-Id: I8f493aaa9228647e08ea5badb06dcfe716d6925d
Reviewed-on: https://go-review.googlesource.com/c/go/+/689336
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Junyang Shao <shaojunyang@google.com>