Inconsistency between KSampler and SamplerCustom from discarding penultimate sigma

### Expected Behavior

With same setting, the result of KSampler and SamplerCustom should be the same. If not, it would need to show a hint to explain the difference at somewhere.

### Actual Behavior

Currently, KSampler implements discarding penultimate sigma for specific samplers (dpm2, unipc) to try to make their result better. ([code here](https://github.com/comfyanonymous/ComfyUI/blob/ab4dd19b913be95dc7f1a8080cd49aa940345c96/comfy/samplers.py#L768))
However, sigmas and sampler for SamplerCustom are created individually, which cannot handle penultimate sigma now.
It leads to different result when using those sampler even with same setting.

### Steps to Reproduce

Make a workflow including KSampler and SamplerCustom with same setting. Use dpm_2, dpm_2_ancestral, uni_pc or uni_pc_bh2 would find result different. Other samplers would be same.

The first image uses UniPC, and second one uses DPM++ 2M to show the difference.
Workflow above: [unipc_workflow.json](https://github.com/user-attachments/files/16620705/unipc_workflow.json)

### Debug Logs

```powershell
No relevant log
```

### Other

_No response_