Text this: M2Caps: learning multi-modal capsules of optical and SAR images for land cover classification