I would like to understand the advantages of the script versioning introduced in SegWit.

As I understand, script versioning allows you to change the meaning of existing opcodes, where currently we can only add opcodes.

Why is this needed? Isn't the power of this simple stack based language that you can add *any* functionality simply by adding new opcodes, (and can do so indefinitely with an OP_EXT), without the need for versioning?

Isn't that a very clean mechanism to update this language while retaining backwards compatibility? Especially because due to the immutability you can never deprecate old functionality.