New Model Grouping hides all non-default quants from the api

Nobody has claimed this yet.

I was previously able to pick what I wanted for any model/quant using the list of returned downloaded models, but after this update, I now only get the default single quantization with no way to see the others.

Ex: I downloaded 3 quants for granite-4-h

lms ls and the python sdk basically return this, with just the first mondel
```
{
"type": "llm",
"modelKey": "ibm/granite-4-h-tiny",
"format": "gguf",
"displayName": "Granite 4 H Tiny",
"publisher": "ibm",
"path": "ibm/granite-4-h-tiny",
"sizeBytes": 4231029813,
"paramsString": "7B",
"architecture": "granitehybrid",
"quantization": {
"name": "Q4_K_M",
"bits": 4
},
"vision": false,
"trainedForToolUse": true,
"maxContextLength": 1048576
},
```

How do I programmatically get to know about the other quants and load them?

Start with the lms ls behavior and the Python SDK's downloaded-model listing, reproducing the case with three Granite 4 H quants. Trace where models are grouped and where only the default quantization is returned. Done means callers can discover and load each downloaded quant through the API, with the listing behavior verified.

Written by the indexing model from the issue text.