[Nanobrowser] Add persistent "Knowledge" store for agents

### What problem does this solve? Currently, when working with agents through the nanobrowser extension, I often find myself repeating the same background context or instructions in prompts. This is inefficient and error-prone, especially for tasks that require domain-specific or user-specific knowledge that doesn’t change frequently. ### Proposed solution I'm proposing a new feature called "Knowledge", which would allow users to persist structured or unstructured pieces of information that can later be reused by their agents when solving tasks. Motivation By introducing a persistent "Knowledge" feature, users will be able to save reusable context that their agents can automatically access when reasoning or executing instructions. Feature Scope ✅ Create a "Knowledge" section in the extension UI where users can view, add, edit, and delete knowledge entries. ✅ Entries could be plain text, Markdown, or structured JSON. ✅ When a task is initiated, the agent should be able to automa