Suggested Enhancement: Multimodal multi turn support

[MoRocety]

As it is, the MNN-LLM App supports multiturn conversation for text only LLMs.

Trying to build the app yourself and by simply enabling history for the app only works to retain text history, not image nor audio. And it keeps referring to the first attachment despite subsequent attachments if attempted to ask questions about the latter.

Tested on MNN-LLM App Version 0.1