Make lab_to_target more parallel for improved in-logic compilation

[myreen]

This issue is about investigating whether the lab_to_target compiler phase can be made faster for in-logic compilation. The current implementation of lab_to_target almost forces computation to be a tedious global process.

Suggestion: one could possibly modify the lab_to_target phase so that a large part of the computation can be done in parallel for each function in the in-logic compilation. One could e.g. compute a maximum encoding length for each function and then assume in other parts that this is the function's exact length. If the encoding turns out to be shorter, then one can add padding at the end to make the function's encoding in order for the actual encoding to always be the maximum encoding length.

Since the lab_to_target phase is complicated and its verification proof is delicate, I suggest trying to make minimal changes before more adventurous changes are attempted.

Note: The lab_to_target compiler must not be changed on master before the install-and-run branch has been merged. Thus any immediate work on this should be a fork off of install-and-run.

[tanyongkiam]

Closed in favor of #1080