Saga pattern helper — distributed transactions with compensations

[pathosDev]

Distributed-transaction-style workflows ("book flight, charge card, email confirmation, on failure roll back") map naturally to actors but writing the orchestration by hand is repetitive. Saga is the standard pattern: a list of steps, each with an explicit compensation action that runs on failure.

API sketch:

```ts
const result = await Saga.run([
{
do: async () => bookingActor.tell({ kind: 'reserve', flightId }),
compensate: async () => bookingActor.tell({ kind: 'release', flightId }),
},
{
do: async () => paymentActor.tell({ kind: 'charge', amount }),
compensate: async () => paymentActor.tell({ kind: 'refund', amount }),
},
{
do: async () => emailActor.tell({ kind: 'send', template: 'confirmation' }),
// no compensation — emails can't be unsent, but step 3 is the last one anyway
},
]);
```

Scope:

• `Saga.run(steps, opts?)`: runs steps in order; on failure, runs `compensate` of all completed steps in reverse.
• `opts.persistenceId?` enables crash-safe sagas — write each step's status to a journal so a restart resumes from where it stopped.
• Built on top of `PersistentActor` when persistenceId is set; pure in-memory otherwise.

Out of scope:

• Cross-process saga coordination (saga lives in one actor; the steps it triggers can be cluster-wide).
• Long-running (> hours) sagas with state that lives in DD — separate issue if needed.

Estimate: 4-5 days.

[pathosDev]

Superseded by #179, which covers the same pattern at wider scope — durable workflow state, a compensation registry, and retry/timeout policies (sized L against this issue's 4–5 days).

The API sketch and the scope notes from this issue have been transferred to #179 (see that comment) so nothing is lost, and #179 now back-references this issue. Closing so the pattern has a single tracking issue rather than two.