fix(mcp): use relative path for httpx download request (#5003)

## Summary

- The download route's `http.get(f"/tasks/{task_id}/result")` has a
leading `/` which causes httpx to replace the base URL's path (RFC
3986), sending requests to `https://host/tasks/…` instead of
`https://host/api/v0/tasks/…`
- Fix: use relative path `tasks/{id}/result` and ensure `base_url` ends
with trailing `/`

Follow-up to #5001 — caught by AI code review before it hit production.

## Test plan

- [x] 60 tests pass (routes, integration, result_store)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Sourced from commit bdcdb323a05f2b27f552d53e136ca6c36e87fd7f

Author: RafaelPo

futuresearch-mcp/src/futuresearch_mcp/routes.py

@@ -272,12 +272,17 @@ async def api_download(request: Request) -> Response:  # noqa: PLR0911
         )
 
     try:
+        # Trailing slash on base_url is required for httpx to append
+        # relative paths correctly (RFC 3986).  A leading slash on the
+        # request path would *replace* the base path, sending the
+        # request to https://host/tasks/… instead of …/api/v0/tasks/….
+        base = settings.futuresearch_api_url.rstrip("/") + "/"
         async with httpx.AsyncClient(
-            base_url=settings.futuresearch_api_url,
+            base_url=base,
             headers={"Authorization": f"Bearer {api_key}"},
         ) as http:
             resp = await http.get(
-                f"/tasks/{task_id}/result",
+                f"tasks/{task_id}/result",
                 params={"offset": 0, "limit": 100000},
             )
             resp.raise_for_status()