Tests: full-surface equivalence and completeness suites for the DecomposedStrategy migration

Two complementary suites pin the migration from the monolith Strategy to
DecomposedStrategy with no relaxations and no vacuous green.

StrategyReferenceSurfaceTests - the SAME completeness checks applied to both
implementations over the same combined deterministic backtest scenario (SMA
trading, 0.2% stop protection, raw tick/level1/depth subscriptions, a forced
off-step register fail with connector CheckSteps, a scripted resting order
that is edited, re-registered, canceled and canceled again, cancel-on-stop,
final Reset). Shared crossover decision and shared order-maintenance script
guarantee identical inputs. For each side: every drivable event must fire at
least once, structurally undrivable events must stay silent (each with the
reason documented), aggregate properties must carry real data (position
moved, PnL/commission/slippage accumulated, collections and statistics
non-empty), and every order/trade must be fully stamped (ids, times, final
states, ownership marks, per-trade commission). The monolith passes except
for genuine reference data gaps it reproduces (the CheckSteps-rejected order
keeps TransactionId=0 and default Time; a registered order keeps default
Time); the decomposed side is red across the board: 35 events never fire,
properties stay zero/null, no entities are produced.

StrategyDecomposedFullEquivalenceTests - strict 1:1 stream equivalence: both
implementations run the same shared SMA-crossover logic in real deterministic
HistoryEmulationConnector backtests (seed 42, fixed ids,
PassThroughMessageChannel, MatchOnTouch, deterministic commission and
slippage), every externally observable event is recorded into normalized
ordered streams (64 families: subscriptions with criteria and lifecycle at
connector and strategy level, states, candles, signals, market-data relays,
connector-boundary order/trade/fail/position/portfolio traffic,
new/registering/registered/changed/canceling/re-registering/edited orders and
all fail relays, mass-cancel and lookup-result relays, trades, all position
surfaces, PnL with realized/unrealized split plus PnLReceived/PnLReceived2,
commission, slippage, latency, errors, infra notifications including
PropertyChanged and CurrentTimeChanged, drawing surface) and compared 1:1 -
per family, then as one interleaved master stream. The compared window is
configure..Reset, and a synthetic "Final" record compares the end-of-run
property state of both sides. Eleven scenarios: storage candles, candles
from ticks, candles from level1, protection on candles and tick-built
candles, forced register-fail, mid-run stop with resting orders, foreign
connector-level order, scripted edit/re-register/cancel flow, LongOnly
trading mode with comment stamping, and raw market-data relays - each with
data-substance guards so empty-vs-empty can never pass, with per-family
count diagnostics on guard failures.

All tests are deterministic across reruns. The intentionally red tests
reproduce genuine product divergences, primarily:
DecomposedStrategy.RegisterOrder/ReRegisterOrder pre-assign
Order.TransactionId, which Connector.CheckOnNew rejects, so on a real
connector the decomposed side registers zero orders (mock-based tests never
exercised this path); DecomposedStrategy attaches a foreign order while it
has no own transactions; no Stopping->Stopped transition in StrategyEngine;
register-fail relays never reach the strategy due to the subscription
CanProcess gate; no order stamping, no IsOnline tracking, no
strategy-position or subscription-scoped events, no edit/cancel-fail/relay
surfaces, dead TradingMode/CommentMode properties, silent Reset and PnL
notification gating differences.

Author: mikasoukhov