Document (on the website) reducer semantics with an eye to transactionality and repeating reducers

[gefjon]

Such a document should cover:

• The 1:1 correspondence between host-to-WASM calls and transactions.
• The fact that a reducer which calls another reducer will not start a sub-TX.
• The fact that a repeat attribute on a reducer (mostly) behaves like putting a schedule!(delay, self(ctx.timestamp)); call at the end of the reducer.
  • Different when called by another reducer? I think yes.
  • Different when the reducer fails? I think no.