Live scaling fails with “unhandled exception” when cpu.corespersocket is set (need support for consistent CPU sockets)

### problem

I’m trying to enforce a consistent CPU socket configuration across VMs by using the global setting cpu.corespersocket.
However, when this setting is enabled, live CPU scaling fails with the message:

`unhandled exception`

If I remove the setting cpu.corespersocket, the same VM live-scales successfully, so the issue is directly related to that parameter.

For some environments (especially those with CPU-based licensing tied to sockets), it is important to make sure that the socket count is always fixed (e.g., always 2 sockets regardless of the number of cores assigned).

At the moment, CloudStack does not seem to reliably honor or handle this setting during live scaling operations.

### versions

Apache CloudStack: 4.22.0
Hypervisor: VMware vCenter + ESXi
Infrastructure: Standard VMware cluster, shared storage
VM type: User instance with dynamic scaling enabled

### The steps to reproduce the bug

1. Set the setting on any VM/instance:
> `cpu.corespersocket = 2`
2. Turn on the VM/instance
3. Scale VM/instance
4. `Unhandled exception` error pops up

### What to do about it?

- Please confirm whether the cpu.corespersocket setting is fully supported for VMware live scaling.
- If this is a bug, kindly update VM reconfigure logic to correctly compute:
`cores = total_vcpu / cpu.corespersocket`
`sockets = cpu.corespersocket`
- If this is a limitation, please advise if CloudStack can support a feature to force a fixed socket count (e.g., always 2 sockets) regardless of vCPU changes.
- This is important for environments that rely on socket-based licensing.

No source file or test is named. Start by reproducing live CPU scaling on VMware with cpu.corespersocket=2, then trace the VMware live-scaling entry point and its VM reconfiguration handling. Done means scaling no longer raises an unhandled exception and the socket/core configuration is either preserved or its limitation is clearly documented.

Written by the indexing model from the issue text.