Multilingual self-supervised speech representations improve the speech recognition of low-resource African languages with codeswitching