googleai: add native ResponseSchema support for structured output

[DragonFSKY]

Summary

Add support for Google AI's native structured output feature using ResponseSchema. This enables users to define JSON schemas that the model will strictly follow when generating responses.

Partially addresses #1342

Changes

• Add ResponseSchema field to llms.CallOptions for structured output support
• Add WithResponseSchema() helper function in llms package
• Export Schema type and Type* constants from googleai package for convenience
• Auto-set ResponseMIMEType to "application/json" when using schema
• Add validation for MIME type compatibility (returns error if incompatible)
• Add comprehensive unit tests for schema types and options

Usage

import (
    "github.com/tmc/langchaingo/llms"
    "github.com/tmc/langchaingo/llms/googleai"
)

schema := &googleai.Schema{
    Type: googleai.TypeObject,
    Properties: map[string]*googleai.Schema{
        "name":  {Type: googleai.TypeString, Description: "User name"},
        "age":   {Type: googleai.TypeInteger, Description: "User age"},
        "roles": {Type: googleai.TypeArray, Items: &googleai.Schema{Type: googleai.TypeString}},
    },
    Required: []string{"name"},
}

resp, err := llm.GenerateContent(ctx, messages,
    llms.WithResponseSchema(schema),
)

Implementation Details

• Uses Google's native ResponseSchema field in GenerationConfig (not prompt-based JSON mode)
• Type-safe: Returns explicit error if wrong schema type is passed
• Backward compatible: No changes to existing API behavior
• Follows the same pattern as existing WithResponseMIMEType() and WithJSONMode() options

Testing

• All existing unit tests pass
• Added new tests for:
  • Schema type aliases and constants
  • WithResponseSchema option functionality
  • Backward compatibility with existing options
  • Complex nested schema structures