Provide agent instruction set on converting incidence to prevalence data

[brandomr]

What's new

This PR addresses the need to convert incidence data to prevalence data. This is accomplished by injecting a set of instructions, defined in incidence_to_prevalence.md to the agent as part of the generate_code tool. I think this is preferable to hardcoding a templated procedure since it will be much more flexible in terms of its ability to handle datasets without super standardized formats and/or novel model compartments.

Risks

The downside to this is that it increases the overall amount of information in the agent's context so therefore increases cost slightly. More importantly though, we don't really know how this new information in the agent's context might impact its decision making / code output for other tasks. This isn't easy to test, but based on my general assessment I think the risk is pretty low and it does seem like this incidence to prevalence conversion is one of if not the most important data transformation tasks.

Testing

This was tested with these datasets on staging:

{"d1":"2fb95ea8-8e71-47c6-b6a9-9c2d6d31484e",
"d2": "06a36516-f4ad-4233-a5fe-dd4a2e3343cd",
"d3":"561b00ad-331f-4d0c-8eb2-0dd813290eef"}

and the following prompt:

Let's assume that average time to recover is 14 days and average time to exit the hospital is 10 days. Can you convert this data into prevalence data? Ideally please map it to SIRHD. Assume a population of 150 million.