nx_regalloc_linscan_test.nx

nx_regalloc_linscan_test.nx -- prove the register allocator on two IRs: A. LOW pressure (a dependency chain, <= 2 live at once): nreg=8 -> 0 spills, memory traffic collapses (the stack tax eliminated). B. HIGH pressure (a tree reduction with a long-lived value, 5 live at once): nreg=3 -> MUST spill, and the allocation MUST STILL BE SOUND (no two live non-spilled vregs share a register -- the exact property the prior g1 regalloc attempt violated). Memory traffic still far below the baseline. Known answer: both allocations VALID (independent checker), A has 0 spills, B has spills > 0, and both reduce memory traffic vs the stack baseline. exit 0.