fix(orders): full-close exit re-expands to grown position under pyramiding

[luisleo526]

Summary

A qty_percent=100 strategy.exit re-issued every bar while the position grows via pyramiding/DCA had its reserved quantity frozen at the size from when it was first placed. clear_existing_exit_order captured the prior order's absolute qty into preserved_reserved_qty, and compute_exit_reserved_qty honored it unconditionally — overriding the 100% intent. At the TP touch only the first FIFO lot closed at the true limit; the residual lots exited one bar late at a re-priced limit, fragmenting one logical exit across two bars and producing wrong exit prices for all but the first lot.

Fix (1 logical line)

Gate the preserved-qty carry to genuine partial (qty_percent < 100) re-issues:

if (!std::isnan(preserved_reserved_qty) && qp_io < 100.0 - kFullPercentEps) {

A re-issued full-close exit now falls through to recompute against the full current position (the prior order is already cleared from pending_orders_), so the whole stack closes at the single TP touch. Partial re-issues keep the carry (no double-reserve); the explicit-qty / from_entry bracket path never calls this function and is unaffected.

Gates

• ctest: 75/75 passed — incl. test_multi_tier_exit_precedence, test_fills_edge, test_accounting_reconciliation; adds test_full_close_while_pyramiding (reverting the gate flips it to 5 failures — the test has teeth).
• Corpus: verify_corpus --all = 251 excellent / 1 documented anomaly — exactly baseline, zero tier regressions.
• Minimal repro (3 pyramid lots, 100% TP): all 30 units close at the TP on the trigger bar (PnL 120), not fragmented across two bars (PnL 200).
• Edge cases reasoned + green: exactly-100% (clamped), >100% (clamped), explicit-qty/OCA brackets, genuine partials, multi-tier precedence.

🤖 Generated with Claude Code