will chose `+++` as section divider between templates

update the parser so it can return parsed prompt templates from the serialized file. for e.g. it should parse the file below -

```
tags: education, physics, simple
---
role: system
source: input
text:
you are a helpful AI --- assistant that +++ explains
\\---complex concepts as if explaining to a 5-yr old
---
role: user
source: input
text: Can you explain how a rainbow is formed?
+++
tags: jinja2
---
role: ai
source: output
text:
{% for color in ['Red', 'Orange'] %}
A rainbow --- is formed by +++ {{color}}.
\+++{% endfor %}
---
role: user
source: input
text: Can you explain jinja2 templates?
```

- as 2 prompt templates
- prompt_template[0].tags = ["education", "physics", "simple"]
- prompt_template[0].prompts[0] = {role: "system", source: "input", text: "\nyou are a helpful AI --- assistant that +++ explains---complex concepts as if explaining to a 5-yr old"}
- prompt_template[0].prompts[1] = {role: "user", source: "input", text: "Can you explain how a rainbow is formed?"}
- prompt_template[1].tags = ["jinja2"]
- prompt_template[1].prompts[0] = {role: "ai", source: "output", text: "\n{% for color in ['Red', 'Orange'] %}\nA rainbow --- is formed by +++ {{color}}.\n+++{% endfor %}"}
- prompt_template[1].prompts[1] = {role: "user", source: "input", text: "Can you explain jinja2 templates?"}
