It is kind of fuzzy. They are not that upfront with milestones, maybe because those are shifting a bit. It looks like it may be Dec / Jan or so. Most of the work is done by Max, so he is doing really a lot to get it off the ground. Therefore, it would be foolish to keep the lack of (hard) milestones against the project as it really goes with the high speed, and quirks are surfacing from time to time. Still, the number of features implemented is really big. Also, what I have seen so far is that any issue that was brought up was addressed really quickly (e.g., today during the dust storm there were plenty of fixes).
Currently all runs on testnet7. Today was a dust storm day, and some new stuff came out (got at least suppressed if not fixed right away). So, maybe one / two more dust storms are still going to happen on tn7. Although, this is the end of tn7.
The early assumption was that tn8 will kick in early next week. That should be the last testnet, as such a month or two after that the mainnet should kick in. Again, this is just my reading of what was said (not following too much lately).
The interesting part is that it looks like every block won on tn8 will transition to mainnet as 0.25 MMX per block. I guess, this is to incentivize people to go full time before mainnet and help to get all kinks out.
Currently, the highest netspace I saw was around 20-30PB. It is expected to be maybe a couple hundred of PB when hitting the mainnet.
So far, there is no plot NFT support. However, it is expected that it will be there day one. That implies that some heavy testing will need to happen on tn8 to get it ready for the mainnet. My read is that the plotter will need to be reworked, as potentially that ~20% plot compression will be there as well. Maybe some extra optimization will be done to get it closer to BB performance. So, that may need to be factored in how long tn8 will last. (Again, my reads of previous statements, not an official one.)
One interesting fact is that the mainnet will support k30 plots and up, and it is expected that those will be good enough for about 3 years. I would assume that Max has an excellent read of what is current or coming as far as number crunching tech. That would mean that k32 plots will potentially be sunset in about 10 years or so (two generation of HDs). So, most likely plot format changes (e.g., plot compressions) will trigger replotting faster than anything else.