Text this: Towards bridging the synthetic-to-real gap in quantitative photoacoustic tomography via unsupervised domain adaptation