But I still have no idea what "composition" in "first frame of composition" means?
It means the full feature including all assets that are assigned to it. First frame to last frame including video, audio, and if present, captions. All that is included in the CPL that makes the composition, it means 'composition play list'. 'Feature' does not just mean 'Full length feature', but any of the composition types - features, trailers, ads, etc. Note that some DCPs, or, distribution packages may contain multiple compositions, CPLs (e.g. multiple language versions or a 2D and 3D composition). Each of these compositions then may contain it's own marker set.
An SPL - Show Play List is created on the cinema server or TMS, and usually includes multiple CPLs placed in a sequence, and with added auditorium control cues for lights, curtains, etc.
Some markers should be present in all SMPTE DCPs (now standard). They are not a strict technical requirement for playback (otherwise DCP-o-matic would create them automatically), means, a server will not complain if markers are not present. Some servers even don't know what to do with markers at all. Maybe DCP-o-matic should or will create the basic markers automatically in the future - makes sense for FFOC and LFOC.