Add strict_types configuration option for JSON Schema type compliance

[sergiobayona]

Summary

Add a strict_types configuration option that enables strict type checking to match JSON Schema specification behavior. When enabled, type validation will reject values that don't exactly match the declared type, even if they could be coerced.

Background

Currently, EasyTalk uses ActiveModel's numericality validation which coerces strings to numbers. This means:

class User
  include EasyTalk::Model
  define_schema do
    property :age, Integer
  end
end

user = User.new(age: "30")
user.valid?  # => true (string "30" is coerced to integer 30)

Per JSON Schema specification, a string "30" should not be valid for type: integer - the type must match exactly.

Proposed Solution

Global Configuration

EasyTalk.configure do |config|
  config.strict_types = true  # default: false (backwards compatible)
end

Per-Model Configuration

class User
  include EasyTalk::Model
  define_schema do
    strict_types true  # override global setting for this model
    property :age, Integer
  end
end

Behavior

Value	Type	strict_types: false	strict_types: true
30	Integer	valid	valid
"30"	Integer	valid (coerced)	invalid
30.0	Integer	valid	valid (per JSON Schema, 30.0 is a valid integer)
30.5	Integer	invalid	invalid
"hello"	Integer	invalid	invalid
3.14	Float	valid	valid
"3.14"	Float	valid (coerced)	invalid
true	Boolean	valid	valid
"true"	Boolean	invalid	invalid

Use Cases

When coercion is helpful (default):

• HTML forms submit everything as strings
• Query params are strings
• CSV imports
• Rails-style "params come as strings" workflows

When strict types matter:

• LLM function calling (OpenAI, Anthropic) - exact type contracts
• API validation where you control the input format
• Data integrity guarantees
• JSON Schema compliance testing/certification

Implementation Notes

• Default should be false for backwards compatibility
• Consider adding per-property override: property :age, Integer, strict: true
• Error messages should be clear: "must be an integer" vs "is not a number"
• Reference: Pydantic has similar strict=True mode

Related

• JSON Schema Test Suite: type.json tests
• Test added in spec/easy_talk/schema_validation_spec.rb documenting current coercion behavior

Labels

enhancement, json-schema-compliance

[sergiobayona]

Related: #146 expands on this to add comprehensive type coercion for all property types (String, Integer, Float, Boolean, Date, Time, etc.)