nx_dwline_e2e_gate.nx

nx_dwline_e2e_gate.nx -- COMPILER -> ASSEMBLER -> ELF -> DECODER, on the real toolchain. nx_dwline_roundtrip_gate proves our DWARF encoder and decoder agree ON A BUFFER. That is a real two-system check but it does NOT prove the shipped pipeline produces such a buffer. This gate drives the actual sovereign toolchain -- nx_cc_sovereign.elf -g, then nxasm_x86_main.elf -- reads the ELF that falls out, and resolves an address through nx_addr2line_lib. GROUND TRUTH IS THE SOURCE FILE, NOT OUR OWN OUTPUT. runtime/nx_probe_dbgtiny.nx defines dbgtiny_a on line 5 and main on line 9. GNU readelf independently resolved 0x400096 -> line 5 and 0x4000AD -> line 9 on this exact probe, so asserting "e_entry resolves to line 9" is checked against an external oracle and against the source, not against a number we made up. THE LOAD-BEARING TOOTH IS THE NEGATIVE CONTROL. Compiling the SAME probe WITHOUT -g must produce an ELF with NO .debug_line at all. Without it, a pipeline that emitted debug info unconditionally would pass every positive tooth here -- and it would silently regress the published "168 B minimal static binary" measurement, which is a MEASURED exceed on /compare/lang. DDR-002 named that tradeoff as its load-bearing risk; this is the check that holds it. license_tier: ORIGINAL expect_exit: 0