nx_panic.nx

nx_panic.nx -- enterprise panic handler. When Nishi-owned code encounters a condition it cannot handle (unreachable branch taken, invariant violated, untrusted input poisoned the parser), the panic handler produces a STRUCTURED post-mortem dump and exits cleanly. What a proper panic produces: * A header line with tag + message (nx_log FATAL level). * Register state -- sp, ra, optionally t0..t6 (future: read via signal handler ucontext_t). * Stack hex dump -- 128 bytes starting at sp (future: walk CFI to unwind frames symbolically). * A structured footer (seq + exit code) so harnesses can correlate the panic to the exit. * Exit with NX_PANIC_EXIT=200, same as nx_assert. Comparison to production systems: * Rust std::panic! with RUST_BACKTRACE=1 -- tag + msg + sym bt. Our v0.0.1 matches on tag+msg; bt is symbolic via future DWARF-native parser. * Go's panic(x) -- dumps stack + goroutine info. Same shape. * Google glog LOG(FATAL) -- msg + crash dump + core. Same shape. * Linux kernel oops -- registers + stack + modules. Our long-term target. The panic handler is ENTRY-LEVEL enterprise today, upgrade path to full-fidelity when: * SIGSEGV handler installed (runtime can catch hardware faults, not just invariant violations) * DWARF symbol table lookup available (symbolic bt) * Record-replay captures state for offline analysis