jobreg-registry-reconcile adjudications file degrades to zero suppression silently, so previously-ruled items return as fresh findings

Found by bin-venus-cc 2026-09-12 while verifying v1.7's header against venus data under the OPS-220 jobreg lift. Hit by accident, and the first reading was wrong in the reassuring direction, which is the point. ADJUDICATED resolves to $(dirname "$(readlink -f "$0")")/jobreg-reconcile-adjudicated.tsv, so a copy run from anywhere but its own directory looks for the rulings file beside itself, finds none, and load_adjudications returns at [[ -r "$ADJUDICATED" ]] || return 0 with NO message. The run then reports '0 adjudicated' and re-surfaces every previously-ruled item as a live finding - bin-venus-cc's first run showed applog-pull@terra.timer that way and it looked like v1.7 having lost the suppression. Re-running with JOBREG_ADJUDICATED pointed at the real file gave the identical-set result. NOT A v1.7 REGRESSION: that default is v1.6's and predates the version bump. What makes it worth a row is the asymmetry v1.7 introduced - v1.7 newly NAMES its jobreg source in the header in all three states (override, fresh, cache-unreachable) and refuses a zero-row manifest, while the adjudications input gets none of that treatment. Same shape as the zero-byte registry v1.6 already guards: an absent input answering confidently in the reassuring direction, where the reassuring answer is 'nothing was suppressed' rather than 'I could not read the rulings'. It bites whenever the script runs from outside its own directory, which is exactly what a diagnostic invocation does. Fix bin-venus-cc proposes and I endorse, cheapest consistent with what v1.7 already does: name the adjudications source in the header alongside the other two, so a reader sees 'adjudications: <path> (N rulings)' or 'NONE LOADED' and cannot mistake the second for the first. Acceptance: a run with the rulings file present prints the path and a nonzero count; a run with JOBREG_ADJUDICATED pointing at a nonexistent path prints NONE LOADED and still exits usefully; and a captured before/after showing the applog-pull@terra.timer case suppressed in the first and reported-with-caveat in the second. HELD UNDER THE FLEET STOP - bin-whey-cc's file and version bump, not to be started until Elazar's lift covers it; the jobreg lift covered OPS-220 only.