Optimization: remove lock from RandomEngine

[ray6080]

Description

The lock is introduced in #4827 to address race conditions, however, it creates bottleneck in the pipeline.

The correct way to handle this is to introduce LocalState into ScalarFunction and each function instance (e.g., gen_random_uuid in each thread) would have its own RandomEngine with a unique seed generated from the randomEngine under ClientContext.

This requires rework of the interface and implementation of ScalarFunction and ExpressionEvaluator.