Add nested array and object support to the schema DSL

[simonw]

Mainly because I want the NY Times demo to return an array of string names of places and people.

Follows:

• Implement a concise schema DSL alternative to JSON schema #790

[simonw]

I came up with this design:

name str, hobbies [name, duration int], address {street, city, zip int}

And got Claude to try implementing it: https://claude.ai/share/f7213da2-9e58-495b-822b-6a8c02c151ae

[simonw]

I won't include this in the first release of the feature. Claude's implementation is too buggy. https://claude.ai/share/f7213da2-9e58-495b-822b-6a8c02c151ae

This example from the Claude generated docs:

llm schemas dsl "name: person's full name
    age int: age in years
    address {
      street: street name and number
      city: city name
      state: state or province
      zip int: postal code
      coordinates {lat float, long float}: geographical location
    }: physical address
    contacts [
      {
        type: contact type (email, phone, etc.)
        value: contact details
        preferred bool: whether this is the preferred contact method
      }
    ]: list of contact methods
    hobbies [str]: list of hobbies and interests
    employment_history [
      {
        company: employer name
        title: job title
        years_active int: duration of employment
        projects [{name, budget float}]: major projects worked on
      }
    ]: work experience"

Produced the wrong result - note how address.street is top level when it should be nested:

{
  "type": "object",
  "properties": {
    "name": {
      "type": "string",
      "description": "person's full name"
    },
    "age": {
      "type": "integer",
      "description": "age in years"
    },
    "address": {
      "type": "string"
    },
    "street": {
      "type": "string",
      "description": "street name and number"
    },
    "city": {
      "type": "string",
      "description": "city name"
    },
    "state": {
      "type": "string",
      "description": "state or province"
    },
    "zip": {
      "type": "integer",
      "description": "postal code"
    },
    "coordinates": {
      "type": "object",
      "properties": {
        "lat": {
          "type": "number"
        },
        "long": {
          "type": "number"
        }
      },
      "required": [
        "lat",
        "long"
      ],
      "description": "geographical location"
    },
    "}": {
      "type": "string"
    },
    "contacts": {
      "type": "string"
    },
    "{": {
      "type": "string"
    },
    "type": {
      "type": "string",
      "description": "contact type (email, phone, etc.)"
    },
    "value": {
      "type": "string",
      "description": "contact details"
    },
    "preferred": {
      "type": "boolean",
      "description": "whether this is the preferred contact method"
    },
    "]": {
      "type": "string",
      "description": "work experience"
    },
    "hobbies": {
      "type": "array",
      "items": {
        "type": "string"
      },
      "description": "list of hobbies and interests"
    },
    "employment_history": {
      "type": "string"
    },
    "company": {
      "type": "string",
      "description": "employer name"
    },
    "title": {
      "type": "string",
      "description": "job title"
    },
    "years_active": {
      "type": "integer",
      "description": "duration of employment"
    },
    "projects": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "{name,": {
            "type": "string"
          }
        },
        "required": [
          "{name,"
        ]
      },
      "description": "major projects worked on"
    }
  },
  "required": [
    "name",
    "age",
    "address",
    "street",
    "city",
    "state",
    "zip",
    "coordinates",
    "}",
    "contacts",
    "{",
    "type",
    "value",
    "preferred",
    "}",
    "]",
    "hobbies",
    "employment_history",
    "{",
    "company",
    "title",
    "years_active",
    "projects",
    "}",
    "]"
  ]
}