Every provider read its response with an unbounded io.ReadAll, or handed
resp.Body straight to json.NewDecoder, so a misconfigured or hostile endpoint
decided how much memory a single search allocated. Two providers made it worse
by reading before they looked at the status code and then interpolating the whole
body into both a log line and the error the caller sees - so a failing endpoint
could push an unbounded blob into the error string as well as into memory.
Reuse the cap the datasource connectors already apply to their API bodies
(readCapped + maxJSONResponseBytes + errResponseTooLarge, added for feishu, ima,
notion and yuque in #3985) rather than inventing a second dialect here. Apply it
to the six providers that read a response body, and bound failed-response
previews the way searxng.go already bounded its own and the connectors bound
theirs.
The two providers that decoded straight from resp.Body now buffer through
readCapped first, since a decoder cannot be capped without buffering.