I don't know if tesla's decided what to do with old cars. They certainly seem to support their roadsters better than, say, VW or audi support their older cars.
Why would tesla switch to LFP for replacement S battery packs? They're battery constrained. A cell that's going into a pack either helps sell a new plaid *or* keeps a crappy 2014 S on the road. Which makes more profit? The Plaid. Why keep the nasty old timers on the road? Marketing.
It's a valid thing for a company to say "go pound sand we're not selling you a part for your car you should throw it out and buy a new one." Someone somewhere has to make the decision of "do we support our older cars because it makes us look good?" (Mercedes and Porsche believe that to be the case).
A strong business case can be made for designing an LFP for the big and small block Ss / Xs can be made. They'll keep the cars on the road, keep the owners who want to keep their car happy, and making a supply of these won't eat into your NMC supply. How much is a project to build a pair of batteries to supply all the small and large packs? Dunno, but it probably won't break the bank.
Lastly, if they wanted to do this, they'd probably just farm the whole task (except firmware) out to BYD. Remember -- the whole point of the S program was to do engineering validation on your customer's dime. Discovering a battery flaw in an enormous program like the 3/Y project would be a company-threatening event while making those same discoveries (battery catches fire? throttle it; degrades excessively? Call it the 90 and move on!) on a program the size of the S project, both with smaller volumes and larger margins, is just good business. Vetting a vendor who wants to sell you new packs by having them make a retrofit LFP pack for legacy S/X cars is just a good vendor hazing project.
So -- I'd hope that tesla has a project to make $5000 / 8000 LFP packs (for 60/70/75 vs 85/90/100) for retrofitting expired S/X packs. Realistically speaking, I'd say there's a 10-30% chance that such a thing will actually be done.