[atol]

What was changed

• Signals and signal-with-start sent through TemporalNexusClient.client in a Temporal operation handler now reuse the Nexus operation's request ID.
• Breaking (experimental API): each handler run can send only one such signal; a second raises a BAD_REQUEST nexusrpc.HandlerError. (Signal-with-start counts toward the limit; a failed signal doesn't; a retried Nexus task starts with a fresh limit.)
• Signals sent through temporalio.nexus.client() are unchanged.
• Based on De-duplicate Nexus Signal and SignalWithStart operations sdk-go#2781.

Why?

When a Nexus task is retried, the handler reruns and sends its signal again with a new random request ID, so the workflow receives it twice. Reusing the operation's request ID lets the server dedupe the retry. However, every signal in one handler run then shares that ID, so a second signal to the same workflow would be silently dropped while reported as delivered. Limiting each run to one signal turns that silent loss into a clear error.

Checklist

1. How was this tested:
   Added integration tests.

2. Any docs updates needed?
   No.

[VegetarianOrc]

Looks good to me! Ty for the changes! Just a couple of nits I'd consider before merging.

[VegetarianOrc]

nit, but I'd prefer this encapsulated in a context manager that manages the signal slot context var. could be a static method on _SignalSlot so we could do like

with _SignalSlot.limit_signals():
    result = await self.start_operation(start_ctx, nexus_client, input)

[atol]

Moving the dedup into _ClientImpl via _signal_request_nexus_context makes it apply to every client used inside _SignalSlot.scope(). .client is again the same object as nexus.client() and removing it isn't needed for the fix, so I restored it.