real time clock drifting off

Supposedly the crystal is 20 PPM accuracy, so 1 second in 50000 - about 2 seconds a day.

Well, I didn't think the caps were required (since the capacitance is so low and the board has some itself), but I just tried now and it did seem to need them - so yes, I'd add some. They're only a few pence each.

You could always add the crystal, try, and add caps if it's not working.

If you have access to a hot air gun, flux, and tweezers then it'll actually be pretty easy to do. I think with a standard soldering iron it's going to be a bit more of a struggle - but you could put the crystal on its side so you can solder the contacts to the pads more easily.

You can try the following code:

(function() {
  console.log("LSEON",peek32(1073887344)&1);
  console.log("LSERDY",peek32(1073887344)&2);
  console.log("LSION",peek32(1073887348)&1);
  console.log("LSIRDY",peek32(1073887348)&2);
  console.log(["NONE","LSE","LSI","HSE"][((peek32(1073887344)&768)>>8)]);
  console.log("RTCEN",!!peek32(1073887344)&32768);
  // RTC subsecond
  console.log("Subsecond:", peek32(0x40002828));
})();

It should report:

LSEON 1
LSERDY 2
LSION 0
LSIRDY 0
LSE
RTCEN 0

When it's working.