I think I spoke too soon when I said synth recording takes latency out of the equation. I now recall that I reported early on that synth recording wasn't being compensated for input latency like external audio recordings. The reply I got was essentially that since a soft synth might be driven by live MIDI input and existing MIDI playback at the same time, it's not possible to apply latency compensation to synth recording because of the way the output of existing MIDI is buffered in advance so that synth audio behaves like recorded audio with regard to playback latency.
But I think this is a little beside the point because the original screenshot showed MIDI being early vs. the timeline, and by
much more than 10ms, which wouldn't be visible at that zoom level. Since synth audio isn't compensated for input latency, we know that synth audio is going to be very late when the buffer is large. But that's a different issue from the relationship of recorded MIDI to the timeline, and this timing isn't evident from your latest screenshots because your timeline is showing milliseconds instead of M:B:T.
Since it's impossible to perform normally with a very high buffer, I recommend you go back to the 10ms RTL, and re-check the timing vs. the timeline and metronome, ignoring the slight discrepancy between the MIDI and Audio due to the lack of audio latency compensation.