refactor: ITransactionWriteRepository — eliminate catch-to-check in command handlers (Phase 3.4)

[josecarlospeer-cloud]

Problem

Every command handler that needed to verify aggregate existence was using exception control flow:

try {
    await eventStore.load(id);
    // exists — conflict
} catch (AggregateRootNotFoundException) {
    // doesn't exist — proceed
}

Violation: Command handlers orchestrate; they should not query to make decisions using exceptions as flow control.

Source: Vernon (IDDD), p. 356. Also M2 in the billing-api evolution roadmap.

Consequence: The pattern is scattered across every command handler, and the EventStore (infrastructure) is injected directly into Application layer code.

---

Fix

ITransactionWriteRepository (Domain):

exists(id: TransactionId): Promise<boolean>;
load(id: TransactionId): Promise<Transaction>;
save(transaction: Transaction): Promise<void>;

EventStoreTransactionRepository (Infrastructure): wraps EventSourcing.EventStore — the catch-to-check lives in exists() exactly once.

All four command handlers (Create, Confirm, Fail, Refund) now inject ITransactionWriteRepository via infrastructure.transaction.writeRepository and use clean readable checks:

if (await this.writeModel.exists(id)) throw new ConflictException(...)
if (!await this.writeModel.exists(id)) throw new NotFoundException(...)

Test double: InMemoryWriteRepository shares the same InMemoryEventStore as the existing write-side test double, so handlers and assertions see the same state.

---

Tests

41/41 pass.