fix: scope strategy.close replacement by callsite

[luisleo526]

What changed

• add a stable six-argument strategy.close runtime path keyed by compiler callsite token
• preserve the five-argument compatibility path as token zero and leave the public C ABI unchanged
• queue distinct syntactic close sites in first-admission source order while repeated evaluations of one site replace in place
• keep per-site reservation and provenance state isolated across bars and reset it across engine lifecycles
• cover all existing bypass paths including ANY, explicit qty and percent, immediate, non-POOC, intrabar COOF, and consumed-C COOF

Evidence

• native TradingView ledgers: XAU close-ledger cube, A-B-A interleaving, and shared-inner-UDF replacement
• exhaustive 2^2 interaction matrix: 4 of 4 cells passed with no early stop; all-off reproduced the defect and all-on matched every native ledger
• supplemental UI-observed overlap probe also matched; it is not presented as a native CSV export
• fresh Release CTest: 114 of 114 passed
• targeted integration: 1173 passed, 0 failed
• calc-on-order-fills: 281 checks passed
• corpus: 252 of 252, with 247 Excellent, 4 Strong, and 1 known anomaly
• full validation board: 416 of 416 completed, 355 Excellent, 46 Strong, 4 Moderate, 2 Weak, 0 downgrades
• six production-strategy tape and mismatch-set digests were byte-identical across all matrix cells; XAU remained Strong, so this runtime correctness milestone is intentionally tier-neutral

Scope boundary

The cross-owner same-ID persistent backing model is regression-covered and preserves the established physical-capacity rule, but this PR does not claim a new native TradingView tape for that separate cross-bar edge.

Landing order

Merge this engine overload first. The dependent codegen PR will emit nonzero source-location tokens only after this PR is green and merged.