Fix start_block_exec nondeterminism

[lacraig2]

#1447 determined that switching callstack_instr to start_block_exec caused nondeterministic behavior.

This PR attempts to resolve that issue.

[AndrewFasano]

Prior to this, if panda_exit_loop was true was the sequence of events something like:

1. An interrupt/exception/etc occurs which should redirect control flow from block B (about to be executed) to block C (some other place)
2. SBE callback triggers for block B
3. SBE callback triggers for block C
4. Code in block C runs (probably multiple blocks with SBE/EBE callbacks firing as exepected)
5. Control flow returns to block B
6. SBE callback triggers again for block B
7. Block B actually runs

And the issue was that this double-SBE callback for block B should have only happened once (and was non-deterministic in when it happened twice)

[AndrewFasano]

Even if this doesn't fix the issue in #1447, is this still worth merging? Or do you want to try tracking it down further and seeing if you find the fix?

[lacraig2]

I'm not sure it really adds anything in the scenario that it doesn't fix #1447.

Or rather, I expect this and something else will completely resolve #1447, but I haven't had a chance to figure out what that is.