fix(orders): close(id) closes logical unclosed qty, not physical lots (grid bots)

[luisleo526]

Problem

strategy.close(id) under the default FIFO close-entries rule summed every physical open lot carrying id to decide the close quantity. TradingView instead closes the quantity entered under id that a prior close(id) has not already targeted — it does not re-sum physical lots.

The two agree when each id maps to one open lot (the common case), but diverge for grid bots that re-use a single entry id across sequential buy/sell cycles:

• the FIFO trade-record drain removes the oldest lot (often a different id), leaving the id-tagged lot physically open, so the next close(id) double-counts it → over-close;
• a TP whose id-lot was already drained finds no physical match and is skipped → under-close.

Net effect on the 3commas grid corpus: wrong trade count + PnL (xrp-grid 143 vs TV 153; exit-price p90 ~18×).

Fix

Add a per-id id_unclosed_qty_ ledger — "entered qty for id minus already-closed-by-close(id) qty for id" — maintained at every entry site (fresh open, pyramid add, raw fill) and consumed by compute_close_target_qty under the FIFO rule only. The ANY rule keeps the physical id-matched path untouched. No new C-ABI export (in-engine state only).

Validation

	before	after
xrp-grid match vs TV	26.7%	99.3% (143/144, px 100%)
pol-grid match vs TV	21.5%	99.1% (338/341, px 100%)
corpus	251 excellent / 1 anomaly	251 / 1 (held, byte-identical)
ctest	76/76	76/76 + new test_strategy_close_reused_id_closes_one_logical_slot (fails pre-fix)

python3 scripts/check_c_abi_runtime.py exits 0 (no symbol change).

🤖 Generated with Claude Code