1 Textbooks did tell you that as an antenna designer, you should try to make s11<-10dB. Textbooks didn't tell you antennas can also work at s11=-1dB. It's just an efficiency loss. At s11=-3dB, efficiency is 50%, at s11=-1dB, efficiency is about 20%. An antenna does not lose signal immediately as s11 goes above -10dB, you just lose some maximum communication distance.
You should aim to make S11 as good as possible. The reason is not that 'the antenna stops working', but that RF power is expensive, and you really don't want to throw much of it away on a bad match. Some high power amplifiers can actually be damaged if they drive a poor S11 antenna, but they tend to be more robust than that in phones where poor S11 is a likelihood.
As you point out, S21 isn't hit that hard by poor S11.
Generally communication systems are built with sufficient link margin that calls get through most of the time. The output power of the phone (often +20dBm or more) suffers many losses from its antenna to the cell site receive antenna, the largest of which is path loss. A few extra dB lost in radiation efficiency simply reduces the maximum length of the path that the tower-phone link can handle. Would you notice if your serving cell lost 10% of its effective radius? Do bear in mind that cell sites are planned assuming people will hold their phones, and weak spots are infilled if service is bad and there's enough traffic to warrant it.
2 Mobile device engineers can add some resistors on antennas to make those antennas not lose s11<-10dB condition in most scenarios in daily use.
No, absolutely not. The cure of absorbing some RF power at this point is worse than the disease. Resistive attenuators are used to mask poor S11 in many measurement scenarios, but's that's when precision and repeatability are needed, rather than power and noise floor.
3 Today's mobile device baseband chips are very powerful. They have the function of VNA. In daily use, the chip frequently tests the antenna and calculates an optimal path to rematch. Then use its variable matching network to match the antenna to make the communication go again.
When I was working a decade or two ago in this field, active antenna tuning was a research topic, but as I understood it, mainly so that antenna hardware could be switched between different frequency bands on the fly, rather than tuned up for best possible S11. Such active tuning would need additional space and detector hardware, maybe it could be worth it to implement? Comments invited from people in the industry now.
Phones do do a lot of clever things in their DSP these days. Multiple antennae can be used to beam-form, which not only 'points' a directional antenna towards the cell site, but can also point it away from strong sources of interference. Links are more often interference rather than noise-floor limited. This gains a few dB of link margin over single antenna operation.
Wideband communications, allowing OFDM, are much more resistant to multipath fading, which gains a few dB of link margin over narrowband systems.
Newer forward error correction codes can operate to within a sniff of the Shannon limit, which gains a few dB of link margin over older codes and uncoded systems.
