Adds firecrawl_scrape, so the agent can read a page it already has the URL for without installing the Firecrawl CLI.

• scrape.ts posts to /v2/scrape with the same key, _opencode integration tag, and abort and timeout handling as Alexandria. It takes url, formats (markdown, html, links, json), jsonPrompt, and onlyMainContent. json is sent as { type: "json", prompt } and is refused before any request when jsonPrompt is missing.
• Markdown, html, links, and json are rendered into the tool output; a target page answering 4xx or 5xx gets a one-line note, and API failures report status and error detail.
• Registered on OpenCode 1 and OpenCode 2 next to Alexandria. Scrapes spend credits, so like Alexandria the tool appears only once Firecrawl is connected, and connecting or disconnecting mid-session adds or removes it.
• README documents the tool; version goes to 1.5.0 and scrape.ts joins the published files.

Testing

• npm run check passes.
• The repo has no test runner, so I checked behaviour with a throwaway script and a mocked fetch (not committed): request body and auth header, json prompt mapping, rendering, the 404 note, API errors with and without a JSON body, the missing jsonPrompt guard, input validation, and that both OpenCode 1 and OpenCode 2 register the tool only with a key.