Castleofargh (and Killcomic) have already answered your main point, Android will resample to the rate supported by the DAC, commonly in Android phones for example that would be 48kHz, although some are capable of 96kHz.
1. Theoretically, 24bit gives 48dB more dynamic range than 16bit. However, that's not the case in practice, there's no difference in audible dynamic range. This of course assumes that the 16bit and 24bit files are the same master but if they're different masters and additional compression has been applied to the 16bit version, then it would measure roughly 5dB lower dynamic range. However, that is NOT evidence 24bit has more dynamic range than 16bit AND as you're NOT playing different masters in this scenario, just converting sample/bit rates of one master, the audible dynamic range will be the same.
2. No, the only "part of it" lost, is the inaudible parts: Above the frequency limit and below the noise floor of human hearing. Also, not resampling at half the sample rate (multiple of 2) could sometimes make an audible difference with very old resamplers, 25 years or so ago but isn't the case today.
3. Countless ABX and double blind tests, over the course of 2 decades or more, have demonstrated no audible difference, which agrees with the objective measured differences being outside the limits of human hearing. The logical conclusion is that there must have been a fault with the ABX test you conducted, for example, you compared different masters (one with added compression for instance).
3b. You're right, you SHOULDN'T argue with this topic here in Sound Science forum, as you are arguing with well established/demonstrated science and you'd need some exceptional evidence to support your claims or be dismissed as a deluded audiophile making false claims!
G