nx_crawl_callee_probe.nx

nx_crawl_callee_probe.nx -- PER-CALLEE VmSize PROBE for the crawler's per-page pipeline (debt 1787082132). The +282 MB/min ingest growth survived the seg-store release-list fix; the named per-page mappings (wc_harvest 8 KiB + olh_scan ~9.5 KiB) explain only ~20 KB/page, so the dominant allocator is in the per-page CALLEES. This organ MEASURES each callee instead of reading it (the debt row's own law: "do NOT hoist by reading"): run the SAME page bytes through each stage N times and diff /proc/self/status VmSize around the loop. Per stage it prints BOTH numbers that matter: warm_kb = VmSize cost of the FIRST call (one-time lazy tables/policy loads -- not a leak) steady per-call = (delta over N calls after warm) / N -- the number that scales with pages/min Stages: H2T (nx_html_to_text), FEED (nx_feed_discover + nx_feed_extract), OLH (olh_scan), FETCH (nx_https_fetch_follow -- only when /tmp/mozilla_certdata.txt exists AND the arg is a url; otherwise that stage prints UNOBSERVABLE and abstains rather than acquitting). argv: nx_crawl_callee_probe <page-file-or-https-url> [iters] (file mode = no network, deterministic) Output: one CALLEE line per stage; the LAST line names the worst steady grower (positional anchor). license_tier: ORIGINAL expect_exit: 0