[test infra] Tile dimension not dense in L1

[rtawfik01]

Currently any test infra test that sweeps num_faces < 4 or face_r_dim < 16 has an issue where tiles are always placed sparsely in L1 (always by offsets of 32x32). Which means for those sweeps, behaviour is technically undefined since LLK unpack and pack always assume dense tiles in L1. Reason this system has worked so far is because output also assumes sparsity in testbench

this behaviour has been found for matmul, tilize, untilize, & datacopy so far, creating this as a master issue for cleanup.

@fvranicTT @amokanTT fyi