Never used Nixie tubes, so I can't comment on that part.
Adding to what thm_w said:
* Yes, this MCU has an internal RTC, so you could save the DS3231, although I'm guessing you chose it for better accuracy, and you'll indeed get better than the internal RTC and an external 32kHz crystal. But if the accuracy/drift you'd get with the internal RTC is adequate for your needs, something to consider to save some parts. This STM32L4 is a very low power MCU, and in the appropriate stop or standby mode, it would draw no more power than then DS3231 - maybe actually less. So you can absolutely use it as a permanent RTC, and wake up the CPU every second (if your clock is displaying seconds) or even every minute if not, update the display, and get back to sleep.
* You don't need all those oscillators. Since you're not using the internal RTC, you can do without the 32kHz crystal. Yes I suppose 32 "MHz" is a typo. The secondary oscillator can't oscillate that fast anyway, it's really made for 32kHz crystals. You can even do without the other oscillator (HSE). Why a full-blown oscillator anyway and not a crystal? (Could be to save a pin on the MCU, but you're not even using all of them...) The system could absolutely do without the two, the internal oscillator would be more than adequate for what the MCU has to do. But, I see you have an USB connection, and USB will be more stable with a crystal - although I've heard people getting USB to work fine only from the internal oscillator. Anyway, USB seems the only reason here to use a crystal. I'd get rid of the secondary oscillator and the 32kHz crystal and would keep only one crystal using the HSE oscillator, not a full external oscillator. That'll be less expensive. I think 12MHz would be adequate to generate the proper internal frequency for USB. The MCU has internal oscillators, so you can use them coupled to timers to wake-up the device even when the HSE is stopped (to save power).
* Not sure what the JTDI signal is for (JTAG interface.) This MCU supports SWD, and the only required signals are SWDIO and SWCLK, NRST ideally as well (so the probe can reset the CPU, required if it's in low-power mode), and SWO for debugging purposes, although this last one is optional.
* Rotary switch: I suppose this is a rotary encoder? I'd suggest doing some analog filtering with capacitors. Sure you can debounce in software, but without this analog filtering, the signals will be very bouncy, and may give you some headaches software-wise. Look that up in any datasheet for a rotary encoders, most of them include a small schematic showing how to do this.
* I also don't think your 'level shifters' will work properly as is.