Investigate Goodbye response behavior

> Sorry to comment on a closed issue, but this inconsistency still exists.
>
> We ran differential tests against all 6 CL clients on a fulu testnet and observed the following Goodbye response behavior:
>
> | Client | Response to incoming Goodbye |
> |------------|---------------------------------------------|
> | Lodestar | Responds with SUCCESS (echoes reason code) |
> | Teku | Stream reset (no response body) |
> | Prysm | Connection closed / EOF |
> | Lighthouse | No response |
> | Nimbus | No response |
> | Grandine | No response |
>
> Lodestar is the only client that responds to an incoming Goodbye. The other 5 clients process it internally (disconnect, apply backoff) but do not write back to the stream.

_Originally posted by @Alleysira in [#2635](https://github.com/ChainSafe/lodestar/issues/2635#issuecomment-4251039050)_

No file or test is named. Start by locating the incoming Goodbye handling entry point in Lodestar's TypeScript consensus-client code, then compare its response behavior with the relevant protocol expectations and the six-client observations in the issue. Done means establishing whether Lodestar's SUCCESS response is correct and documenting the resulting behavior or required change.

Written by the indexing model from the issue text.