[torchtune integration] post training + eval (#670)

## What does this PR do?

- Add related Apis in experimental-post-training template to enable eval
on the finetuned checkpoint in the template
- A small bug fix on meta reference eval
- A small error handle improvement on post training 


## Test Plan
From client side issued an E2E post training request
https://github.com/meta-llama/llama-stack-client-python/pull/70 and get
eval results successfully

<img width="1315" alt="Screenshot 2024-12-20 at 12 06 59 PM"
src="https://github.com/user-attachments/assets/a09bd524-59ae-490c-908f-2e36ccf27c0a"
/>

llama_stack/providers/inline/eval/meta_reference/eval.py

@@ -15,7 +15,7 @@ from llama_stack.apis.agents import Agents
 from llama_stack.apis.datasetio import DatasetIO
 from llama_stack.apis.datasets import Datasets
 from llama_stack.apis.eval_tasks import EvalTask
-from llama_stack.apis.inference import Inference
+from llama_stack.apis.inference import Inference, UserMessage
 from llama_stack.apis.scoring import Scoring
 from llama_stack.providers.datatypes import EvalTasksProtocolPrivate
 from llama_stack.providers.utils.kvstore import kvstore_impl

llama_stack/providers/inline/post_training/torchtune/recipes/lora_finetuning_single_device.py

@@ -110,6 +110,10 @@ class LoraFinetuningSingleDevice:
             self.checkpoint_dir = config.checkpoint_dir
         else:
             model = resolve_model(self.model_id)
+            if model is None:
+                raise ValueError(
+                    f"{self.model_id} not found. Your model id should be in the llama models SKU list"
+                )
             self.checkpoint_dir = model_checkpoint_dir(model)
 
         self._output_dir = str(DEFAULT_CHECKPOINT_DIR)