Shutdown error logging behaviour

Hi,

we are seeing a lot of error logging messages on v2.46.0 shutdown.

The most common messages,

> 54][filter]Failed to call check, error: CANCELLED: Request cancelled, str body:
> Failed in Check call: Request cancelled

disappear if we remove

```
- --health_check_grpc_backend
- --healthz=healthz
```

however the rest of the logging is a confusing mix of JSON / text / info / warn / error / critical logging. Surely this is not the intended behaviour?

Log extracts:


then a lot of repeated lines


Note about the health checks: the downstream container is still alive when this happens (delayed by a `preStop` sleep for 30 seconds).

No source files or tests are named. Reproduce the v2.46.0 shutdown with --health_check_grpc_backend and --healthz=healthz, then compare the logs with those flags removed while the downstream remains alive. Done should mean shutdown produces consistent, non-repeated logging at appropriate levels.

Written by the indexing model from the issue text.