nx_bench_fetch.nx

nx_bench_fetch.nx -- THE GENERIC SOVEREIGN URL->FILE FETCHER (rule 15 DRY; eats the hardcoded-URL fetcher-family duplication class). WHY IT EXISTS: the ecosystem has ~119 `*_research_fetch` organs, each a copy of the same TLS+save body with its URLs BAKED IN. I wrote two more this session (swebv, gmmlu) before the third instance made the pattern obvious -- rule 15 says extract at 3. Every one of those copies also inherited the same latent bugs (the ff_putn no-write class, the missing self-mkdir) precisely because they were copies. This organ is the one body they should all have called: URL and destination are ARGUMENTS, not source edits. nx_bench_fetch <url> <outpath> [force] outpath MUST be under knowledge/fetched/ or /tmp/ (boundary defense, rule 12) idempotent: an existing outpath is a HAVE-SKIP unless `force` is passed prints: status=<http> bytes=<n> SAVED | HAVE-SKIP | FETCH-FAIL | SAVE-FAIL | GZIP-skip exit 0 saved-or-skipped | 2 usage/denied | 3 fetch failed SELF-MKDIRS knowledge/fetched (rule 20: a tool that needs a dir makes it -- the missing-dir bug that silently SAVE-FAIL'd 119 organs, seq191). license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0