batch: bug in handle key if is null

[vagetablechicken]

https://github.com/4paradigm/OpenMLDB/blob/908e9edbe05a86dccdbf8d1579bad88454480d59/java/openmldb-batch/src/main/scala/com/_4paradigm/openmldb/batch/nodes/WindowAggPlan.scala#L424-L425

We get key by extractKey, and then check it by isValidOrder. After #1408 fixed, isValidOrder arg is java Long, it may be null or a long.

But extractKey returns scala Long, which can't be null, ref https://github.com/4paradigm/OpenMLDB/blob/8a20b4849c52dfee7dacd02f04a6adede05673c2/java/openmldb-batch/src/main/scala/com/_4paradigm/openmldb/batch/window/WindowComputer.scala#L225-L227 The function getLongFromIndex called by extractKey, returns java Long. https://github.com/4paradigm/OpenMLDB/blob/2e7fb76aca6a41a96c43b46b1be2194281c2166f/java/openmldb-batch/src/main/scala/com/_4paradigm/openmldb/batch/utils/SparkRowUtil.scala#L66

We miss null in extractKey. And after I temporarily make extractKey return java Long, some batch tests failed. e.g. https://github.com/4paradigm/OpenMLDB/blob/836b34119249ff9b667130b125c88395ef18ee4f/java/openmldb-batch/src/test/scala/com/_4paradigm/openmldb/batch/nulldata/TestWindowWithNullData.scala#L28

reproduce:

1. make extractKey return java Long
2. run TestWindowWithNullData

[lumianph]

@vagetablechicken this is fixed already?

[vagetablechicken]

@vagetablechicken this is fixed already?

nope. isValidOrder has been fixed, but extractKey is not. And the tests will failed if we fix the extractKey, needs @tobegit3hub to check it.