Fix Postgres column stats UnboundLocalError on errors

[ianu82]

Summary

• return early from meta_get_column_statistics when native_query does not return a table response
• only run dataframe post-processing when a table dataframe is present
• add a regression unit test that verifies non-table responses are returned unchanged

Why

The previous implementation always referenced df after the conditional transformation block, causing an UnboundLocalError when native_query returned an error/non-table response.

Validation

• python3 -m py_compile mindsdb/integrations/handlers/postgres_handler/postgres_handler.py tests/unit/handlers/test_postgres.py
• pytest -q tests/unit/handlers/test_postgres.py -k meta_get_column_statistics_returns_non_table_response (fails in this environment due to missing optional dependency psycopg)

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[ianu82]

recheck