docs: show grouping subscribers under the FastAPI OutboxRouter

[lesnik512]

Refs #158.

Summary

#158 asked whether faststream_outbox.fastapi.OutboxRouter should forward routers= to the inner broker. It should not, and the capability it would provide already exists through router.include_router(...). This PR documents and tests that path.

Why not forward routers=

• FastStream's StreamRouter.include_router accepts a broker-level router, wraps each of its subscribers with the FastAPI compatibility decorator, then calls broker.include_router. That is FastStream's documented "multiple routers" pattern for FastAPI.
• OutboxBroker(routers=...) just calls include_routers in the constructor, skipping that wrap. Under the FastAPI router apply_types=False, so nothing injects arguments: a handler with d: str = Depends(dep) receives the raw OutboxMessage as its body and the Depends(...) object itself as d. No error, wrong values.
• None of FastStream's own FastAPI routers (Kafka, Rabbit, Redis, NATS, Confluent) accepts routers either.

The two open questions in #158 are settled by observation on the include_router path: nested subscribers belong to the same broker, so they start with it in the FastAPI lifespan, and their channels appear in the /asyncapi document.

Changes

• docs/usage/fastapi.md: new "Grouping subscribers" section showing a plain faststream_outbox.OutboxRouter included into the FastAPI router, plus the two shapes that do not work (nesting FastAPI routers raises TypeError; router.broker.include_router skips the bridge). The routers bullet under "What's intentionally not exposed" now gives the reason and points to the section.
• tests/test_fastapi.py: test_included_broker_router_subscriber_resolves_fastapi_depends (INVARIANT) checks that Depends resolves in a nested subscriber and the channel is in the AsyncAPI document. Switching the include to router.broker.include_router(sub) makes it fail.

Checks

• just lint-ci: pass
• pytest --no-cov: 518 passed, 107 skipped (Postgres integration tests; not run locally, no SQL touched)
• just docs-build (mkdocs build --strict): pass

[github-actions]

Benchmark gate

✅ gate passed

scenario	msg/s	delete/msg	WALrec/msg	WALB/msg	fpi	upd	del	dead_tup
consumer/w1/b10	1067	1.000	6.79	916	242	5000	5000	10000
consumer/w1/b100	1066	1.000	6.82	919	243	5000	5000	10000
consumer/w2/b10	1221	1.000	6.80	963	242	5000	5000	10000
consumer/w2/b100	1448	1.000	6.73	954	243	5000	5000	10000
consumer/w4/b10	1132	1.000	6.82	995	261	5000	5000	10000
consumer/w4/b100	1719	1.000	6.87	1039	244	5000	5000	10000
consumer/w1/b100/tfbs100	6688	0.010	6.07	1116	243	5000	5000	10000
producer/w1/b100	2428	0.000	3.04	584	0	0	0	0

Gated (fails the build): delete_calls + tuple counters (upd/del/ins) + the producer's insert_calls, exact; select_calls within +2; wal_records within a 10% band. msg/s, WAL bytes and total calls are informational (timing/FPI noise).