Should not sleep Fixes

[Cyberboss]

Fixes #355

Also splits SHOULD_NOT_SLEEP and SHOULD_BE_PURE tests and adds a minor improvement to the test helpers in that it shows more info where it wouldn't when panicing

[Cyberboss]

As it stands, this is still too slow and uses way too much memory. More improvements incoming.

[Cyberboss]

Okay @SpaceManiac This is latest on /tg/. The 7228 is the number of procs infected with SHOULD_NOT_SLEEP to begin with. Each print-out represents 10s of work.

7228 procs to anaylze call chains for
serialized objtree in 1.718s
11/7228 procs anaylzed
165/7228 procs anaylzed
449/7228 procs anaylzed
818/7228 procs anaylzed
1273/7228 procs anaylzed
1775/7228 procs anaylzed
2250/7228 procs anaylzed
2792/7228 procs anaylzed
3313/7228 procs anaylzed
3840/7228 procs anaylzed
4339/7228 procs anaylzed
4800/7228 procs anaylzed
5349/7228 procs anaylzed
6045/7228 procs anaylzed
6610/7228 procs anaylzed
7109/7228 procs anaylzed
dreamchecker 181.200s - total 192.546s

At this point the only optimization I can think of is to NEVER revisit procs (currently procs that don't have a sleep somewhere in their calls can get revisited) It would result in a slightly incomplete error set but it might be worth it.

[Cyberboss]

I'm going to go ahead and do that

[Cyberboss]

Back under 5s. It no longer prints out transitive errors but the time save is worth it.

[Cyberboss]

There's a bug where the checking in /New() is too aggressive (We KNOW the type at that point and shouldn't be checking overrides).

[Cyberboss]

/tg/ circa now dreamchecker 2.882s - total 18.013s ~136 errors

[Cyberboss]

If anyone wants to take over this PR. It's basically done, except you need to get the commented out test from the above commit working