fix: preserve flat all-in entry transaction sizing

[luisleo526]

Summary

• Preserve the placement-quantized quantity for one narrowly scoped, omitted-quantity PoE=100 pure STOP placed from true flat when it becomes marketable at the immediately following bar open. Margin admission and dispatch consume the same quantity.
• Finalize an exact two-call, distinct-ID, opposite omitted-quantity MARKET book from true flat at the next broker boundary and reject the later call only when the gross transaction exceeds placement equity.
• Keep all other order shapes on established behavior through explicit fences for scheduling mode, position state, costs, sizing, pyramiding, OCA, risk, replacement/cancellation, extra calls, delayed/intrabar fills, and configuration changes.

Why

The target chronology contains two independent broker-emulator defects that mask one another. An exhaustive, route-locked N=2 matrix shows neither change alone reaches parity; only both produce exact 481/481/481 canonical and 591/591/591 raw strict+price-exact trades. A second exhaustive N=2 matrix on an independent community strategy shows MARKET alone reaches exact canonical count while STOP is byte-inert.

Evidence

• Aht N=2: 4/4 exhaustive, publication gate true; only all-on is exact; final tape 2827a9eb9243ac476e0c923cbb9f4fb5c9245f8bf1a44c4fa3326cd5c3c620bb.
• CNTV N=2: 4/4 exhaustive, publication gate true; 00 == 10 and 01 == 11; MARKET tape 5e19626ac7b296084f69af23547bbfe4d029c8f36c53b6e553504ab3487f9e93.
• Fresh public-base causal corpus: base and candidate each ran 252/252 successfully against identical feed/generated-source provenance; all tapes and canonical rows are identical, with zero causal downgrades.
• Forced-no-cache scraper gate: all 412 rows accounted, 2 UP / 0 DOWN; candidate board 346 Excellent / 48 Strong / 3 Moderate / 5 Weak plus 8 excluded. Only Aht and CNTV move Strong to Excellent.
• Focused assertions: MARKET 163/163; STOP 78/78; derived-feed self-tests 8/8.
• Full engine CTest: 110/110.
• Fresh independent release review: PASS; neither factor is approved for separate release.

Identity

• Base: 98260ae39fd04e0ea3aacbde5cf65f2bd43e4d8d
• Tested head: 930b67e38000e4317b05f77e4ce9dc6c3e253be3
• Tested tree: d710529c54887c6028b4d0cb11d6ae8ca95ec35e
• Full patch SHA-256: d05bb85e38c99dc14311c8c9731dc8f1e818e8b4146366d4e2b9f17cc72591a6
• Engine library SHA-256: 95abec688bc8013424f13e5f6e97b9b0c1ebbe8d9d136779012c876a4a6ec925
• Corpus comparison SHA-256: 52ca91a95d435c658e03954d0f49d52f38624ca88f11e3baedcace1dbdcdb838

Scope

No Pine source, TradingView tape, OHLCV feed, verifier routing, tier rubric, corpus pointer, or codegen behavior changes.