1) What is the best solution using linear regulators? The classics are two regulators (but then have to add balancing resistors) or bypass transistor. Which is best and why? Any parts/designs suggestions?
Current sharing between multiple regulators can also be enforced with current shunts on their inputs and operational amplifiers to control the additional regulators. This has the advantage of perfect current sharing and not compromising the load regulation with ballast resistors.
However some regulators however have an output voltage tolerance good enough to use low values of ballast resistors and this would be the simplest way.
2) Any other topologies/solutions?
A discrete solution with transistor, operational amplifier, and reference is eminently feasible but gives up the inherent protection features of an integrated regulator. Or an integrated regulator can be used in place of the power transistor for its built in protection features; see below about lower noise.
Some low dropout integrated regulators work down to 0.6 volts so could drive an external power transistor with a total dropout below 2 volts. Most low dropout regulators use a Sziklai pair for higher efficiency but these have a dropout of 1.2 volts which would be marginal when driving an external power transistor.
You mentioned the importance of noise. An external reference and operational amplifier can control a regulator to produce much lower noise than the regulator alone. This also makes remote sense practical.
3) Any other concerns? For example, I'm thinking the 48W dissipated inside an enclosure without ventilation? The elements that heat up (regulator, bypass transistor etc) would have heat sinks but how safe is that?
Power is a real problem in a sealed enclosure. I would mount the heat sinks on the outside of the case and mount the power devices to them through the enclosure.
Integrated regulators have limited power dissipation and that is the criteria which will require multiple regulators in parallel, which is just as well because it spreads the heat out making it easier to dissipate. However operating the regulators at a fraction of their individual current limits means the total current limit will be much higher than the design current which may be undesirable. So some consideration should be given to implementing an external current limit.
Integrated regulators also have secondary breakdown protection which requires them to be used in parallel at high input to output voltages.
4) Would any potting (of the entire PCB) improve or worsen the thermal situation and general safety?
Potting could improve the situation however mounting the power packages to the case would be better and sufficient.