MAINT Breaking: Message Normalizer Refactor

[rlundeen2]

This PR refactors the MessageNormalizer abstraction to standardize how PyRIT converts Message objects into formats required by different LLM targets or situations. Or, as an example, to translate Message to a string that could be understood in a system prompt. It also adds documentation and improves test coverage.

This is an update to the old ChatMessageNormalizer class. This also refactors AzureMLTarget which made use of this, and removed other legacy places that used ChatMessage.

Note: this is part of a bigger change to normalize prepended_conversations for attacks. To see a prototype of that, see d647054

[romanlutz]

We're removing an entire module that users might have interacted with so this is most definitely BREAKING.

[rlundeen2]

We're removing an entire module that users might have interacted with so this is most definitely BREAKING.

For us only used in ChatMLTarget, and it was broken. So I doubt folks were using it, but worth marking anyway