Support for Spark 3.3.1

[da-tubi]

• Python 3.9 Spark 3.2.1 -> Python 3.9 Spark 3.3.0
• Scala 2.13 Java 8 Spark 3.2.0 -> Scala 2.13 Java 8 Spark 3.3.0
• Scala 2.13 Java 11 Spark 3.2.0 -> Scala 2.13 Java 8 Spark 3.3.0

Why Spark 3.3.0 should be tested with Python 3.9?

Databricks 11.3 LTS is using Spark 3.3.0 and Python 3.9

[Renkai]

Consider cache the LD_LIBRARY_PATH? The default behavior of cache:pip only caches very limited content.

https://github.com/eto-ai/rikai/actions/runs/3494328797/jobs/5850014468#step:10:8

[da-liii]

Well, this pr has passed the Python CI. Let us improve CI speed in another PR.

[da-liii]

Two CI failed, because it failed to download the pretrained model from torchvsion.