Fix low-impact potential race condition

[zplizzi]

Consider this execution order:

• User calls this script, and the sync_send line runs to completion
• User separately changes the code and then calls a command to sync code. That code runs to completion
• The send line here executes

That means that the NEW code will be executed, ie, the execute request is not atomic.

In practice it's probably not a huge deal, but it seems like it would both be safer and simpler if we had only a single request type, and that request executed atomically (either to update the code and then execute the cell, in the case or this script, or to just update the code, in the case of the other script)