feat: Add retrieve online documents v2 method into pgvector

[YassinNouh21]

What this PR does / why we need it:

This PR enhances the PostgreSQL online store to support hybrid search capabilities, combining both vector similarity search and full-text search.

Specifically:

• Introduces the ability to perform hybrid queries using both embeddings and keyword-based search.
• Extends retrieve_online_documents_v2 function to handle vector-only, text-only, and hybrid cases gracefully.
• Improves feature retrieval by dynamically selecting features based on query type (distance, text_rank).
• Adds comprehensive integration tests to validate:
  • Vector similarity search (L2 and cosine distance)
  • Full-text search
  • Hybrid search (vector + text)
  • Edge cases (non-matching queries, category filtering)

This update supports the broader goal of enabling more intelligent, contextual document retrieval in Feast's online stores.

Which issue(s) this PR fixes:

Fixes #5115
Part of the roadmap to Introduce Feast NLP/LLM Add-On, enabling advanced search capabilities in vector databases.

Misc

[ntkathole]

cur.execute and cur.fetchall() is repeated in all conditions.

if hybrid search, params = [embedding, tsquery_str, string_fields, tsquery_str]
if vector search, params = [embedding],
.....

cur.execute(query, params)
rows = cur.fetchall()

[ntkathole]

can this be simplified ?

def sort_key(item: Dict[str, Any]) -> float:
            return item["vector_distance"] if embedding else item["text_rank"]

[YassinNouh21]

@ntkathole can u take a quick look

[franciscojavierarceo]

Maybe there's a more explicit way to handle this? Feels like this could be cleaner.

[YassinNouh21]

do u think this will be a better version

has_string_features = any(
                    f.dtype.to_value_type() == ValueType.STRING 
                    for f in table.features
                )

[franciscojavierarceo]

yes!

[YassinNouh21]

@franciscojavierarceo done take a look

[ntkathole]

Looks good to me!

[YassinNouh21]

@franciscojavierarceo I think the reason behind the failed of the ci is this
from line 222 to line 225
at this file in sdk/python/tests/integration/online_store/test_universal_online.py

 sdk/python/tests/integration/online_store/test_universal_online.py
         # writes to online store via datasource (dataframe_source) materialization
         fs.materialize(
-            start_date=datetime.datetime.now() - timedelta(hours=12),
+            start_date=datetime.now() - timedelta(hours=12),
             end_date=_utc_now(),
         )

because it is irrelevant to the pr changed files

[franciscojavierarceo]

i believe this is what broke the integration tests

it's unfortunate naming but previously we imported datetime the package and now you've imported the datetime module from datetime the package, which is what leads to the issue.

[franciscojavierarceo]

thanks for this! need to change the import or update the integration test

[YassinNouh21]

@franciscojavierarceo we are ok to merge

[franciscojavierarceo]

🚀🚀🚀