Use krun class, avoid stderr

[ehildenb]

This is an attempt to switch the KAVM class over to using KRun class functionalities instead of manually calling krun.

Blocked on: https://github.com/runtimeverification/avm-semantics/pull/196 Blocked on: https://github.com/runtimeverification/avm-semantics/pull/189 Blocked on: https://github.com/runtimeverification/pyk/pull/105 Blocked on: https://github.com/runtimeverification/pyk/pull/104

Currently, we're seeing a slowdown using this technique:

Typically, I should be able to run each simulation in ~140-200ms, but now we're seeing ~700-1100ms. Most of the time is wasted in Kore -> Kast conversion in the pyk code. To improve this (by about 500ms I estimate), we should do:

• Implement get_cell over Pattern instead of having to convert the whole term to Kast before selecting a cell.
• If that is still not fast enough, implement kore_json_to_dict instead of kjson_to_dict, to avoid having to convert to Kast at all.

[geo2a]

This design is definitely better, but we need to work on performance, as outlined. Thanks @ehildenb!

[geo2a]

@ehildenb I've cherry-picked most of your commits into #222, closing this one.