interactive.do_user_command and prompts should return resulting env

interactive.do_user_command take an Env, and perform evaluation (possibly updating env), but then throw it away. This means that if Mars is called with both -c and -i, the computation performed in the -c command is a) wasted (caching will be undone) and b) unusable (bound variables will be unbound).

Have both of these functions (most importantly, do_user_command) return an env, so the state is preserved. It may be necessary to thread and return more than just an env (an entire interactive state, so as to keep the localtable for example). Therefore maybe it's better to just add a new function to interactive which performs (optionally) both the do_user_command and prompts, preserving state between them.