Apollo MCP Server

# [1.3.0] - 2025-12-10 ## 🚀 Features ### Set destructiveHint in tool annotations - @DaleSeo PR #510 This PR explicitly sets the `destructiveHint` annotation for MCP tools based on the GraphQL operation type. We now set `destructiveHint: false` for queries and `destructiveHint: true` for mutations to avoid relying on client-side spec compliance. We currently only sets `readOnlyHint` based on whether the operation is a query and omits `destructiveHint`. Per [the MCP spec](https://modelcontextprotocol.io/legacy/concepts/tools#available-tool-annotations), `destructiveHint` defaults to true when omitted. The spec also says `destructiveHint` should be ignored when `readOnlyHint` is true, but OpenAI doesn't appear to implement this correctly. ## 🐛 Fixes ### Fix broken non nullable return types in minified schema - @esilverm PR #514 Fix handling of minified return types to support non-null lists and other non-null cases to improve operation construction accuracy. ## 🛠 Maintenance ### Clean up sanitize module - @DaleSeo PR #503 Just cleaning up unused code ### Abstract away operation execution logic from the server's running state - @DaleSeo PR #517 I abstracted the operation execution logic from the server's running state, following the pattern used in apps. This change helped me write tests and identify a subtle bug where the execute tool wasn't propagating the OTel context.