* separate execution endpoint
  * pros:
    * easy to see how many hits are coming in to execute vs just compile
    * can potentially separate endpoint to different servers, rate limit, queue separately?
    * maybe cache better as filters are unimportant (if cache at all)
    * better separation of concerns
    * ultimately allows for execute-only UI?
  * cons:
    * needs two POSTs in the client (probably not a problem)
    * maybe complicates the UI? Toggle code, code/execute, execute?

POST /api/execute/<lang>/<compiler>
{ code: "", stdin: "", libraries, compile options } as per compile?

On disk compilation cache? could then separate stdin from compile? S3 put? md5sum? retention policy? redis? will that end up costing too much?

Execution queue?