"Framework" and "runtime" get used interchangeably in AI agent conversations, but they answer different questions. A framework answers how do I write an agent? A runtime answers where does my agent run, and what keeps it working once it's live? The difference shows up