fix: streamline theme extraction from overlaps in MultiHopSpecificQue… (#2347)

## Issue Link / Problem Description
<!-- Link to related issue or describe the problem this PR solves -->
- Fixes #2275 
- I also encounter this problem on my test on multi hop specific query
synthesizer:
1 validation error for ThemesPersonasInput
themes.0
Input should be a valid string [type=string_type,
input_value=['Vedenjäähdytyskoneen', 'Vedenjäähdytyskone'],
input_type=list]
For further information visit
https://errors.pydantic.dev/2.11/v/string_type

- @czhiming-maker have not update it already 2 weeks past, so I might
give a try to update on it.


## Changes Made
<!-- Describe what you changed and why -->
- add helper function _extract_themes_from_overlaps from the dicsussion
in #2275
- 
- 

## Testing
<!-- Describe how this should be tested -->
### How to Test

```
from ragas.testset import TestsetGenerator
from langchain_community.document_loaders import DirectoryLoader, TextLoader
from langchain_openai import AzureOpenAIEmbeddings, AzureChatOpenAI
from ragas.testset.synthesizers.multi_hop.specific import (
    MultiHopSpecificQuerySynthesizer,
)

import os

loader = DirectoryLoader("./data/", glob="**/*.md", loader_cls=TextLoader)
documents = loader.load()

# Set your Azure OpenAI credentials
AZURE_ENDPOINT = os.environ.get("AZURE_OPENAI_ENDPOINT", )
AZURE_API_KEY = os.environ.get("AZURE_OPENAI_API_KEY", "")
AZURE_API_VERSION = "2024-12-01-preview" 
AZURE_DEPLOYMENT = os.environ.get("AZURE_OPENAI_DEPLOYMENT", "gpt-4o-mini")
EMBEDDING_DEPLOYMENT = "text-embedding-ada-002"

# Initialize embeddings
embeddings = AzureOpenAIEmbeddings(
    azure_endpoint=AZURE_ENDPOINT,
    api_key=AZURE_API_KEY,
    api_version=AZURE_API_VERSION,
    azure_deployment=EMBEDDING_DEPLOYMENT
)

# Initialize LLM with JSON mode enabled
llm = AzureChatOpenAI(
    azure_endpoint=AZURE_ENDPOINT,
    api_key=AZURE_API_KEY,
    openai_api_version=AZURE_API_VERSION,
    azure_deployment=AZURE_DEPLOYMENT,
    temperature=0.3,
    model_kwargs={
        "response_format": {"type": "json_object"}  # Force clean JSON output
    }
)

generator_llm = LangchainLLMWrapper(llm)
generator_embeddings = LangchainEmbeddingsWrapper(embeddings)

distribution = [
    (MultiHopSpecificQuerySynthesizer(llm=generator_llm), 1),
]
generator = TestsetGenerator(llm=generator_llm, embedding_model=generator_embeddings)

async def generate():
    # generate testset
    testset = generator.generate_with_langchain_docs(
        documents,
        testset_size=4, 
        query_distribution=distribution,
    )

    testset.to_evaluation_dataset().to_jsonl("testset.jsonl")

import asyncio
asyncio.run(generate())
```

## References
<!-- Link to related issues, discussions, forums, or external resources
-->
- Related issues: #2275 


## Screenshots/Examples (if applicable)
<!-- Add screenshots or code examples showing the change -->
Samples Generation in test
<img width="1701" height="111" alt="image"
src="https://github.com/user-attachments/assets/890d14bc-cd31-4940-8ce3-df8bc5ea459b"
/>

---
<!-- 
Thank you for contributing to Ragas! 
Please fill out the sections above as completely as possible.
The more information you provide, the faster your PR can be reviewed and
merged.
-->

Co-authored-by: Kenzo Yan <kenzo.yan@granlund.fi>

Author: kenzoyan

src/ragas/testset/synthesizers/multi_hop/specific.py

@@ -87,11 +87,8 @@ async def _generate_scenarios(
                     ):
                         logger.debug("Overlapped items are not strings or iterables.")
                         continue
-                    themes = (
-                        list(overlapped_items.keys())
-                        if isinstance(overlapped_items, dict)
-                        else overlapped_items
-                    )
+                    themes = self._extract_themes_from_overlaps(overlapped_items)
+
                     prompt_input = ThemesPersonasInput(
                         themes=themes, personas=persona_list
                     )
@@ -100,10 +97,9 @@ async def _generate_scenarios(
                             data=prompt_input, llm=self.llm, callbacks=callbacks
                         )
                     )
-                    combinations = [
-                        [item] if isinstance(item, str) else list(item)
-                        for item in themes
-                    ]
+
+                    combinations = [[theme] for theme in themes]
+
                     base_scenarios = self.prepare_combinations(
                         [node_a, node_b],
                         combinations,
@@ -117,3 +113,30 @@ async def _generate_scenarios(
                     scenarios.extend(base_scenarios)
 
         return scenarios
+
+    def _extract_themes_from_overlaps(self, overlapped_items: t.Any) -> t.List[str]:
+        """
+        Extract unique entity names from overlapped items.
+
+        Handles multiple formats:
+        - List[Tuple[str, str]]: Entity pairs from overlap detection
+        - List[str]: Direct entity names
+        - Dict[str, Any]: Keys as entity names
+        """
+        if isinstance(overlapped_items, dict):
+            return list(overlapped_items.keys())
+
+        if not isinstance(overlapped_items, list):
+            return []
+
+        unique_entities = set()
+        for item in overlapped_items:
+            if isinstance(item, tuple):
+                # Extract both entities from the pair
+                for entity in item:
+                    if isinstance(entity, str):
+                        unique_entities.add(entity)
+            elif isinstance(item, str):
+                unique_entities.add(item)
+
+        return list(unique_entities)