millis_to_timestamp(CAST(created_on AS bigint)) returns fewer rows than raw created_on column in SQL query

[PEPravin]

When querying a Druid datasource via PyDruid, I noticed that applying millis_to_timestamp(CAST(created_on AS bigint)) on a column ingested as a timestamp leads to fewer rows being returned than querying the raw column directly.

Environment:
Druid version: 30.0.1
PyDruid version: 0.6.9
Python version: 3.8.8
Datasource ingestion type: Kafka ingestion with timestamp transformation

Ingestion spec snippet:

{
  "type": "expression",
  "name": "created_on",
  "expression": "timestamp_parse(created_on, 'yyyy-MM-dd HH:mm:ss.SSS')"
}

**Sample raw value:**
"created_on": "2025-08-08 03:45:07.009"

Query and Python code:

from pydruid.db import connect
import pandas as pd

def druid_runner():
    conn = connect(
        host='<host>',
        port=443,
        path='/druid/v2/sql/',
        scheme='https'
    )
    curs = conn.cursor()

    # Query 1: Using millis_to_timestamp + CAST
    sql_query_ts = """
    SELECT  millis_to_timestamp(CAST(created_on AS bigint)) AS created_on
    FROM <datasource_name>
    WHERE opCode <> 'D'
    """
    df_ts = pd.DataFrame(curs.execute(sql_query_ts))
    print("Rows with timestamp conversion:", len(df_ts))

    # Query 2: Selecting raw column
    sql_query_raw = """
    SELECT created_on
    FROM <datasource_name>
    WHERE opCode <> 'D'
    """
    df_raw = pd.DataFrame(curs.execute(sql_query_raw))
    print("Rows with raw created_on:", len(df_raw))

if __name__ == '__main__':
    druid_runner()

**Observed result**
Query with timestamp conversion → 74,945 rows
Query with raw created_on → 101,332 rows

Expected both queries to return the same number of rows.