Trade-by-Trade Realized P&L Engine With FIFO Matching

You are building a real-time P&L engine for a trading desk. A stream of $m$ timestamped trades arrives, each represented as a pair $(p_j, q_j)$ where $p_j$ is the execution price and $q_j$ is the signed quantity ($q_j > 0$ for buys, $q_j < 0$ for sells). Your engine must process the stream in a single pass using $O(m)$ time and $O(1)$ extra space (beyond the inventory queue), and return three things: 1. **Realized P&L** -- computed using FIFO (first-in, first-out) inventory matching. When a trade offsets an existing position, match it against the oldest inventory lot first. 2. **Ending inventory and average cost** -- the remaining open position after all trades, along with its weighted-average cost basis. 3. **Maximum drawdown of cumulative realized P&L** -- the largest peak-to-trough decline in the running realized P&L over the trade sequence. Your implementation must handle the following edge cases: - **Partial fills**: a single trade may only partially offset the oldest lot, leaving a residual. - **Zero-quantity records**: trades with $q_j = 0$ (no-ops that should be skipped). - **Fees**: each trade may carry an optional fee that reduces realized P&L. - **NaN prices or quantities**: invalid records that should be skipped with a warning. **Constraints:** -