Fully leverage LLM APIs to allow users to specify different roles in custom prompts

[wan-robert]

Current Functionality

User writes some custom prompt in a .md file. All texts that the user writes are sent to the LLM with the "user" role.

Desired Functionality

LLM APIs allow users speak for the system, the user, and the assistant at the start of the conversation. In some literature, this is known as the "pre-fill". This helps LLMs retrieve the correct knowledge for the conversation and helps set the tone of the LLM's reponse.

Obsidian Copilot chat histories are already stored in a conversation-like format, so I guess a lot of the code to realize this feature already exists.

For example, if the user writes the following text in a custom prompt

**system:** This is a custom system prompt. It sets the role of the assistant.
**user:** This is a user message in the first paragraph. It sets the tone of the conversation.
This is a user message in the second paragraph. It allows the user to enter more than one paragraph when writing the "user" message for the prompt.
**ai:** The user writes text here to speak for the LLM. This helps set the tone of the assistant response in the actual conversation.
**user:** The user can keep writing more custom prompts as the "user" role.
**ai:** The user can also keep writing more custom prompts as the "assistant" role.

When the user calls this prompt, it will be sent to the LLM in a format similiar to this:

content: "This is a custom system prompt. It sets the role of the assistant."
role: "system"

content: "This is a user message in the first paragraph. It sets the tone of the conversation.\n
This is a user message in the second paragraph. It allows the user to enter more than one paragraph when writing the "user" message for the prompt."
role: "user"

content: "The user writes text here to speak for the LLM. This helps set the tone of the assistant response in the actual conversation."
role: "assistant"

content: "The user can keep writing more custom prompts as the 'user' role."
role: "user"

content: "The user can also keep writing more custom prompts as the 'assistant' role."
role: "assistant"