Use closures instead of genfun?

Your compile logic currently uses [generate-function](https://github.com/mafintosh/generate-function) to build some code, e.g. https://github.com/mafintosh/protocol-buffers/blob/012e982de02db16693ac777816677cbf6f38697e/compile.js#L105.

This is neat (it looks like in some cases you are dynamically including/excluding entire sections of code!) — however it does mean that it won't work on the current [Tessel](https://tessel.io/) runtime, where JavaScript code is actually compiled "offline" before it is pushed, and later `eval`/`Function` is [not currently possible](https://github.com/tessel/runtime/issues/220#issuecomment-49558098).

I know this is just one platform, and it's not really your fault that its "JavaScript" implementation is incomplete, but this was kind of a bummer. Protocol buffers seem like a great fit there. (FWIW, you may also run into this restriction in-browser with default [Content Security Policy](https://developer.chrome.com/extensions/contentSecurityPolicy#JSEval) settings in affect.)

Anyway, just thought I'd mention this for future consideration. I know performance is a big focus here, but seems like simply returning closures instead of "generated functions" would bring compatibility (and readability) benefits, and [speculatively] might not be appreciably slower on modern optimizing JS engines.

Start in compile.js around line 105 and inspect how generate-function builds the protocol-buffer code. Compare the generated-function approach with the proposed closures, including the Tessel runtime and browser Content Security Policy constraints. Done means the compatibility and readability trade-offs are resolved and the compile path no longer depends on runtime eval or Function if that approach is accepted.

Written by the indexing model from the issue text.