fix(trogon-gateway): restore server_info().max_payload once async-nats race is resolved

[yordis]

Background

trogon_nats::connect always sets retry_on_initial_connect, which causes async-nats to return the Client before the background connection task has populated the ServerInfo watch channel. Calling server_info().max_payload immediately after connect() therefore returns 0 — the Default value for usize — not the actual server limit.

MaxPayload::from_server_limit(0) saturates to a threshold of 0, routing every payload through the object-store claim-check path regardless of size.

Root cause in async-nats

async-nats maintains two separate sources of truth for max_payload:

1. An Arc<AtomicUsize> initialized to 1024 * 1024 — used internally for publish size enforcement, correctly updated on connection
2. A ServerInfo watch channel seeded with ServerInfo::default() (where max_payload = 0) when retry_on_initial_connect is set — this is what server_info().max_payload reads

The bug is that the watch channel seed uses Default::default() instead of the same 1 MiB default the Arc uses. async-nats even comments line 1017: "We're setting it to the default server payload size" — they applied the correct default to the Arc but not to the ServerInfo seed.

Current workaround

TROGON_GATEWAY_NATS_MAX_PAYLOAD_BYTES (default 1_048_576) — see PR #119.

What to restore when fixed

// In crates/trogon-gateway/src/main.rs
let server_max_payload = nats.server_info().max_payload;
let max_payload = MaxPayload::from_server_limit(server_max_payload);

Remove nats_max_payload_bytes from GatewayConfig / ResolvedConfig / config.rs and the TROGON_GATEWAY_NATS_MAX_PAYLOAD_BYTES env var.