Add `cache_control` field to `ToolDefinitionJSON`

[mdrxy]

The OpenRouter API supports cache_control on tool definitions for Anthropic prompt caching (announced in Tool Caching), but the Python SDK's ToolDefinitionJSON model only has function and type fields.

When a dict containing cache_control is passed through the SDK's serialization layer, the field is silently stripped.

import openrouter

client = openrouter.OpenRouter(api_key="...")
client.chat.send(
    model="anthropic/claude-sonnet-4-5",
    messages=[...],
    tools=[
        {
            "type": "function",
            "function": {"name": "search", "description": "Search the web", "parameters": {...}},
            "cache_control": {"type": "ephemeral"},  # <-- stripped by SDK before reaching API
        }
    ],
)

The SDK correctly models cache_control on ChatMessageContentItemText (message-level caching works), but the same field is missing from ToolDefinitionJSON.

Expected behavior

cache_control should be preserved on tool definitions and sent to the API, matching the behavior already implemented for message content items.

Suggested fix

Add an optional cache_control field to ToolDefinitionJSON (or its OpenAPI source schema), consistent with ChatMessageContentItemText.cache_control:

class ToolDefinitionJSON(BaseModel):
    function: ToolDefinitionJSONFunction
    TYPE: Literal["function"] = "function"
    cache_control: Optional[ChatMessageContentItemCacheControl] = None  # <-- add this

SDK version

openrouter==0.6.0

Additional context

• The OpenRouter API spec at /docs/api/api-reference/chat/send-chat-completion-request does not currently include cache_control on ToolDefinitionJson either, so this likely needs an OpenAPI spec update before SDK regeneration.
• ChatMessageContentItemCacheControl (with type: "ephemeral" and optional ttl) already exists in the SDK and can be reused.

[ibbybuilds]

following this