Fix spurious error logs from Valkey connection monitor during graceful shutdown

## Problem Description

When shutting down the manager server, spurious ERROR level logs are emitted from the Valkey connection monitor task:

```
ERROR ai.backend.common.clients.valkey_client.client Error in Valkey connection monitor:
Traceback (most recent call last):
File ".../client.py", line 536, in _monitor_connection
await asyncio.sleep(_DEFAULT_MONITOR_INTERVAL)
asyncio.exceptions.CancelledError
```

This is followed by timeout warnings from aiotools:

```
WARNING aiotools.server Timeout during waiting for child processes; killing all
WARNING aiotools.fork Force-killed hanging child: 88863
```

## Root Cause

In `MonitoringValkeyClient._monitor_connection()`, the exception handler catches `BaseException` which includes `asyncio.CancelledError`. During graceful shutdown:

1. aiotools server framework propagates cancellation to all tasks
1. The `_monitor_connection` task receives `CancelledError` before `disconnect()` is called
1. At this point, `self._closed` is still `False`
1. The error is logged even though cancellation is a normal shutdown scenario

## Solution

- Handle `asyncio.CancelledError` explicitly and re-raise without logging
- Change `except BaseException` to `except Exception` to exclude `CancelledError`
- Fix a bug in the finally block log message format

## Files Changed

- `src/ai/backend/common/clients/valkey_client/client.py`

JIRA Issue: BA-3593

Start in src/ai/backend/common/clients/valkey_client/client.py at MonitoringValkeyClient._monitor_connection(). Review the cancellation path and the finally-block log formatting, then exercise graceful manager-server shutdown. Done means normal task cancellation no longer emits the spurious ERROR log or causes the reported shutdown warnings.

Written by the indexing model from the issue text.