Always been in theory possible
In practice this is really really hard the longer the cable, the more variable the cable quality is, connectors and the wider the bandwidth.
Speech on copper pair uses cancellation and there is significant crosstalk. Gigabit ethernet uses this technique and as a result needs much better cable and connections than 100M bps, as well as strict distance limits. it uses all four pairs in parallel, in duplex, versus original cat5 Ethernet which actually only used one separate pair in each direction.
I imagine only part of the 110MHz to 862MHz band will be used and often only on cable that is good quality (no nips, sharp bends, corrosion, poor connections and accurate impedance). At the minute 5MHz to 65 MHz (or sometimes 88 or even 110MHz) is used for upstream which is lower QAM and and TDMA. That's 0.8, 1.6, 3.2, or 6.4MHz channels with virtual TDMA channels in them. Downstream is 2, 4, 6 or 8MHz channels between 110MHz and 560 (old cable), 862MHz or even up to 1200MHz, essentially using DVB-C, but multiplexed data rather than TV.
The problem also is that downstream is really One to Many broadcast (though modems ignore data for other users) and upstream is Many to One, with potentially only 10 simultaneous users if you dedicate a physical channel rather than TDMA. Non-HFC is very much longer cables shared to more users. Hence the typical 8:1 or worse ratio Download/Upload with DOCSIS, though DSL is often worse. Also the QAM is typically lower to reduce modem TX power, this means lower speed. The symmetrical idea would mean very much higher modem power.
A simpler solution on cables not passing too many houses is to use 5 to 500 MHz upstream and 550MHz to 1200MHz downstream. Much more robust than full duplex on the same channels.