online feature didn't change when using kafka to sink data to table

**Bug Description**
openmldb single machine mode
feature: online feature deployment
datastream: using kafka to sink data to table
problems: when sink new data to the table from kafka, I expected feature could be changed, but it didn't happen, feature is old however.
using curl method to get feature: curl http://127.0.0.1:9080/dbs/kafka_test/deployments/demo_data_service -X POST -d'{"input": [[11, 2, 22, 1.2, 1.3, true,"c888", "2022-07-24",1751051906000]]}'

but use "load data into" cmd, feature can change as I expected.

**Expected Behavior**
when sink new data to the table from kafka, feature change too.

**Relation Case**

**Steps to Reproduce**

1.
2.
3.
4.

The issue involves OpenMLDB's online feature deployment and Kafka data sinking. Start by examining the Kafka sink implementation and how table updates trigger feature recomputation. Look for integration points between the Kafka consumer and the feature serving layer. Reproduce the bug by setting up a single-machine OpenMLDB with Kafka, then compare the behavior of 'LOAD DATA' versus Kafka sink.

Written by the indexing model from the issue text.