As suggested, an NTC using a bipolar transistor as a voltage reference + comparator is probably the cheapest solution but then you have the problem of the variability of Vbe v Ic between individual transistors. This isn't usually specified but the BC549 is one that does (min, typ, max @ IC = 2mA): 580, 660, 700mV; you could design to 640mV +/-60mV or approx +/-10% which is approximately +/- 2.5C for a typical NTC so you might just be able to make it work with all the other tolerances.
At least you don't have to worry about the -2mV/C temperature dependence of Vbe as you are only concerned with operation at 0C. In reality though I would expect that the Vbe spread would be much lower than the barn door specs - the manufacturers aren't going to provide tight specs for parameters that are of little interest to the vast majority of customers. So you could do your own batch testing and characterisation to give you enough confidence that this solution will work.
A downside is that you have to prevent the purchasing function from sourcing 'equivalent' transistors from the cheapest supplier, but the chances are that Vbe will differ markedly between different manufacturers of transistors with similar part numbers.
For more precision you could use a TL431 type shunt regulator. These have a 2.5V reference and can be configured as a comparator. They are cheap < $0.01 for Chinese versions and around $0.03 for Diodes Inc's AS431.
That said, I'm concerned about your requirement to charge LIPOs below 0C. Conventional wisdom is that this is a no-no as it can lead to Lithium plating of the anode damaging the battery and creating safety concerns. There are some suggestions that charging at -30C is possible providing you limit charge current to less than 0.02C but I haven't seen anything authoritive - Battery Unversity doesn't count as 'authoritive'. Your requirement is to reduce charge from 50mA to 15mA below 0C - does this mean 50mA is only (50/15*0.02) = .07C? That seems somewhat low (15 hours) but not unreasonable. Surely it would be better to use an NTC to reduce charging current gradually below 0C rather than one of two charge rates? This would at least remove the need for a comparator and reference voltage.
More to the point, does the LIPO manufacturer approve your 'down to -30C' charging scheme, especially given the huge 5C tolerance on the 0C threshold? Your very low cost requirement implies large volume production so liability issues must be a big concern. At the very least you might want to ensure the tolerance is -0C to +5C.
This response to sub-zero charging on stack exchange sems to cover the issue. Apologies if you know all this...
https://electronics.stackexchange.com/questions/263036/why-charging-li-ion-batteries-in-cold-temperatures-would-harm-them