fix: pin_map ranking by observables, not hid

[chouswei]

Problem

pin_map ranking in 0.19.x used hidden Record.hid (_elN) as the sort key. Label sets matched under CREATE-order shuffle; emitted sequences did not. Hid never leaked into the body, but admission order (Lost in the Middle) followed physical CREATE order. That is a gauge anomaly.

Verified in this tree (not only the 0.19.3 install hypothesis):

• bounded_match_find sorted hits by r.hid
• MemStore.list_records / _edge_records / neighbors BFS / context_pack used hid order
• apply_shell_soft_cap and session-outline exemplars sorted by hid (and leftover nickname id)

Fix (0.19.c honesty)

Not a usage-method b, not 0.20, not 1.0. Same operators: Recall = pin_map, Commit = mutate. Hid still off the wire. Package version left at 0.19.3 (c bump is for publish).

• Ranking key = kind + remaining observable payload (exclude hid; exclude nickname id — ranking by nickname is the same class of leak)
• Edges rank by type + endpoint observables
• BFS expansion, fan-out clamp, max_rows admission, shell cap, outline, Peak_L tie-break, CueConflict listing follow that key
• SysML: MN-REQ-04.11 / MN-VER-04-S05 — order of pin_map is a function of the Shape’s observables, not hid

Test

tests/test_pin_map_observable_rank.py builds isomorphic sessions (shuffled CREATE order and shuffled nickname ids), cues by kind + locator slug= (never leftover id), and asserts canonicalised pin_map sequences match. Also asserts _el / _memnet_hid stay off the emit.

Proof on this branch: full pytest green (skips are live Agens/Neo4j). Two sessions with reversed CREATE and shuffled nicknames emit the same canonical sequence; hids differ (_el1… vs _el8…) and do not appear on the wire.

Honesty

• Do not claim 1.0
• N-server stays Later
• No cabinet vendored
• Do not raise (M)