Significant performance regression of signal-slot in heavily multithreaded context

### What happened?

This is due to introduction of `AObjectBase::SIGNAL_SLOT_GLOBAL_SYNC`. If I comment out these locks, the regression dissapears.

### Relevant log output or stack trace

```shell

```

### Steps to reproduce

```bash
Spam AThreadPool with tasks which create AObjects, make connections and emit signals.
```

### Affected Platforms

_No response_

### Link to build, i.e. failing GitHub Action job (if any).

_No response_

Start by profiling the signal-slot path around AObjectBase::SIGNAL_SLOT_GLOBAL_SYNC while reproducing the AThreadPool workload that creates AObjects, connects them, and emits signals. Compare the locked and lock-disabled cases, then define done as eliminating or substantially reducing the measured regression without compromising thread safety.

Written by the indexing model from the issue text.