document the issue of introducer SPOF

[zooko]

Many people learning about the Tahoe architecture for the first time are concerned about introducers as a Single Point of Failure/Single? Point of Load. Most recently kpreid.

Ideally, we would implement #68 -- "decentralize introduction" -- and this would eliminate their concern. Until then, we should explain in the docs:

1. Yes, this is a SPOF/SPOL.
2. But, if the introducer dies, only a minor trouble ensures: newcomers can't get themselves connected.

2.b. And if the introducer can be revived, with the same private key and IP address or domain name, then those newcomers will proceed to get themselves connected, no problem. 2.c. And if it can't be so revived, then a new introducer can be set up, and every node which has its $BASEDIR/introducer.furl file updated to point to the new introducer will start to use the new introducer.

3. We have a plan to decentralize introduction: ticket #68

[warner]

add text to architecture.txt

[warner]

Done, in c83d8b7a6ddf4f8e. I'm not sure if source:docs/architecture.txt will be obvious enough for everybody, but it's googleable.