perf: lazy source traceback in transaction errors

[antazoey]

What I did

I had a dream last night that this was the last remaining performance problem.
Basically all call-based Vm errors seems to calculate their source tbs when they don't always have to.
I am hoping by lazy-ifying this, we re-gain some of performance loss from 0.6.22

How I did it

How to verify it

Checklist

• All changes are completed
• New test cases have been added
• Documentation has been updated

[antazoey]

One example of the performance gain (I am sure there are other cases):

Code:

%time cont.setNumber(5, sender=acct, raise_on_revert=False)

Before:

CPU times: user 264 ms, sys: 8.98 ms, total: 273 ms
Wall time: 391 ms
Out[14]: <Receipt 0x07ea2a17817d0be7191df511223b8d1a2d306390a9c0d93642d59ca026d5dcee>

After:

CPU times: user 209 ms, sys: 7.28 ms, total: 217 ms
Wall time: 322 ms
Out[5]: <Receipt 0x4ea4be20d1949ee762e9f331dd9598c213397e01e13f33bb0c4bd7d37893305c>

^ In this case it's because we are not calculated the ape-traceback because it isnt being raised, because of the raise_on_revert=False flag