Docs suggestion: brief description of how instructor works

[mbforbes]

Hi there, I was reading the readme and the docs, but I couldn't find a concise explanation for how instructor actually works!

I was evaluating whether to use instructor, and wanted to know whether it used constrained decoding, prompting, tool APIs, or something else.

I think it'd be really helpful to have 1–2 sentences in the readme and docs that says something like:

Here's how instructor gets LLMs to produce responses that follow the provided schema:

• If an LLM exposes a structured endpoint, we use that
• Otherwise, if an LLM exposes a tool call endpoint, we use that
• Otherwise, we fall back to a prompt to ask the model to output a response of the provided schema. If it fails, we retry N times
• We don't implement constrained decoding for LLMs that produce logits, but may implement that in the future

(I don't know if the above is accurate!)

As a user of the library, knowing these details are really helpful to understand whether the schema is guaranteed by the LLM in a single call, or whether it's best-effort. Linking to the prompt would be helpful too.

Many thanks for your consideration, and thank you for making and open-sourcing this library!

Is your feature request related to a problem? Please describe.
Understanding instructor's approach.

Describe the solution you'd like
Brief documentation update

Describe alternatives you've considered
Reading the code to figure it out (it's what I did!)

Additional context
n/a

[psharma28]

In my experience, I noticed that the process kept defaulting back to the prompt to give a structured output, which resulted in a Pydantic errors.

I opened up this issue: #1341