document known scaling issues

[zooko]

There are some known issues with scaling Tahoe-LAFS up in terms of servers, users, files, aggregate data. Most of these issues were learned from the allmydata.com experience. Most of them are "soft limits" -- things start taking an inconveniently long time -- rather than hard limits where something that used to work fails.

Examples:

• it took a long time to do deep-add-lease and deep-repair
• user data was spread over all the servers and some files and directories were reachable only through other directories, so if a few servers were (even temporarily) off-line then some users accounts couldn't be deep-add-leased, deep-repaired, or recursively downloaded. There was no good support for a deep- operation to stop cleanly when it encounters an unrecoverable file or directory and show its partial results to the user or preserve its state for later retry.