The conversation history character count calculation is inaccurate

As noted in #1169 the `calculate_char_count` method on `ConversationState` is inaccurate since it is not being performed on the result of `as_sendable_conversation_state`, which is the actual conversation state being sent to the backend.

To facilitate an accurate character count calculation, we need to decouple our in-app representation of the conversation from the data being sent to the backend.

One approach would be to update `as_sendable_conversation_state` to be idempotent. Any code that already calls `as_sendable_conversation_state` would therefore not need to be updated.