TODO

* a noun is an atom or a cell. DONE
* mugs. DONE
* unifying equals. DONE
* dorky embedded DSL for nouns. DONE
* some kind of real hashmap (needed for memoization at least). DONE
* nock->js compiler (THE REALLY INTERESTING PART). DONE (at least, v1)
* decrement. DONE
* port random-noun tests to tape-check. DONE
* jam/cue for pill reading (makes testing easier). DONE
* add. DONE
* jets (basic binding). DONE
* ackerman. DONE
* atom-gates.pill
* ALL jets (ref: jaque)
* boot ivory pill
^--- call urbit library functions from js! great success!
* boot fakezod (brass pill) to dojo
* prevalence (on-disk in node, IndexedDB in brower)
* expose basic vere-slave interface as a library
  i.e. send me pokes, i handle all persistence and send you io 
^--- candidate vere backend
     urbit in the browser
* urbit over webrtc?

At this point, we have a working vere backend, as well as a full
"urbit in the browser".  Performance is unlikely to be as good as jaque,
but will likely be quite accepatable (especially for in-browser things).

At the very least, using an ivory pill to compile and run hoon from the
browser will ease client-side programming tasks.

--- future implementation ideas

it would be feasible to add profiling code to the compiled javascript,
specifically in callsites (9), so that the 9 would run say 100 times and
then recompile itself with a check and a direct call to the most called
battery. this would also enable loop finding in tail position where the
most called battery is the one being compiled. (important here not to
use the 2 compiler cache, but to have a generated call site per battery)

it would be cool to profile that single optimization, all other things 
being equal. it should be the most important one.