In practice you lose more over time so it greatly depends on how many you start with. However, you could encode the DNA in another format which effectively means unlimited shelf life assuming you a reasonably close species to work with.
Another format would actually be a problem - finding a format that would be readable in 10K years is a huge challenge, since all current languages, semantic systems, etc. may be completely forgotten by then. Check out this one:
Persionaly, I suspect a scale model of a DNA strand + it's encoding would be a strong enough hint for any society capable of using that data to build a life form. As up you only need 4 symbols for the encoding. Then just repeat that at ever smaller scales.
The real issue is you need to encode a lot of data in a stable form for a long time for cheap enough someone would pay for it. Pluss, the cost of actually sequencing 10K+ plants * a few examples so you have some genetic diversity.
Probably not. Seeds are live plants enclosed in several layers of protection to make them transportable and to avoid drying out. There are two reasons seeds don't germinate: (1) they are viable (still alive), but it's difficult to break dormancy -- to "wake up" the plant inside the seed and get it to start growing, and (2) the plant tissue inside the seed has died. Seeds that have been stored a long time usually don't germinate because of (2).
Tissue culture requires tissue that is alive. The live embryo is removed from the seed and placed on sterile media containing nutrients and hormones. The exact composition varies depend on the plant species / cultivar you are working with.
Tissue culture is useful as a last resort for breaking seed dormancy, but if the problem is that the plant tissue inside the seed has died, it will be no help.
http://en.wikipedia.org/wiki/Human_Interference_Task_Force
about trying to solve how to say "dangerous radiation" so people in distant future would understand. Not trivial task at all.