fix(openai): stop truncating a provider error before it is parsed

statusError capped the body at 2 KiB and then unmarshalled, so any error longer
than that failed to parse and the caller received a raw JSON fragment cut
mid-object instead of the provider's message. Azure content-filter and
structured-output responses, which echo the schema, routinely exceed it.

The cap moves to 64 KiB: still bounded, but above the largest error body these
providers send. The existing cap test is written against the constant and keeps
holding.

PAGI-132

llms/openai/internal/openaiclient/completions.go

@@ -60,7 +60,7 @@ type errorMessage struct {
 	} `json:"error"`
 }
 
-const maxErrorBodyBytes = 2048
+const maxErrorBodyBytes = 64 << 10
 
 func statusError(statusCode int, body io.Reader) error {
 	msg := fmt.Sprintf("API returned unexpected status code: %d", statusCode)

llms/openai/internal/openaiclient/statuserror_test.go

@@ -30,3 +30,24 @@ func TestStatusErrorCapsTheQuotedBody(t *testing.T) {
 		t.Fatalf("error grew to %d bytes, want it capped near %d", len(err.Error()), maxErrorBodyBytes)
 	}
 }
+
+func TestStatusErrorParsesABodyLargerThanTwoKilobytes(t *testing.T) {
+	padding := strings.Repeat("d", 8<<10)
+	body := `{"error":{"message":"content filter triggered","type":"invalid_request_error","detail":"` +
+		padding + `"}}`
+
+	err := statusError(400, strings.NewReader(body))
+	if err == nil {
+		t.Fatal("want an error")
+	}
+	got := err.Error()
+	if !strings.Contains(got, "content filter triggered") {
+		t.Errorf("provider message lost: %s", got)
+	}
+	if strings.Contains(got, `{"error"`) {
+		t.Errorf("body was quoted raw instead of parsed: %.120s", got)
+	}
+	if strings.Contains(got, padding[:64]) {
+		t.Errorf("padding leaked into the error: %.120s", got)
+	}
+}