fix: Portfolio NAV snapshot understates total_net_gain by ~3x (#397)

[MDUYN]

Summary

Closes #397

Fixes portfolio NAV snapshot total_net_gain being understated by ~3x due to two bugs in the trade service's sell order accounting.

Root Cause

Bug 1 (Primary): Accumulated cost double-counting

In create_order_metadata_with_trade_context, when a sell order closes multiple trades, the cost variable accumulated across loop iterations but the entire accumulated cost was subtracted from each trade's revenue:

# BEFORE (buggy):
cost = 0
net_gain = 0
for metadata in order_metadatas:
    cost += trade.open_price * metadata.amount
    net_gain += (sell_price * metadata.amount) - cost  # subtracts ALL prior costs again

With N trades, earlier costs were subtracted N times → systematic understatement proportional to the number of overlapping trades.

Bug 2 (Secondary): Wrong amount in per-trade net_gain

In _create_trade_metadata_with_sell_order_and_trades, sell_amount (total sell order) was used instead of trade_data["amount"] (per-trade portion) in the net_gain calculation, overcounting gain per trade when a sell order was split across trades.

Fix

# AFTER (fixed):
cost = 0
net_gain = 0
for metadata in order_metadatas:
    trade_cost = trade.open_price * metadata.amount
    cost += trade_cost
    net_gain += (sell_price * metadata.amount) - trade_cost  # only THIS trade's cost

The cost accumulator remains correct for position.cost -= cost below. Only the net_gain calculation changes.

Files Changed

• investing_algorithm_framework/services/trade_service/trade_service.py — Two fixes (per-iteration cost, per-trade amount)
• tests/services/test_trade_service_net_gain.py (new) — 3 regression tests

Tests

All 161 app tests + 451 service/scenario tests + 3 new regression tests pass.