This supercedes #4972 and closes #4613

Notable changes to #4972

• Source code attached to the computation object is now inferred by walking back the stack. All calls originating from dask and distributed modules are ignored. First frame of a non-ignored module is interpreted as user code and we'll extract that source. The ignore list is currently only dask and distributed but we can add other common libs like xarray, prefect as suggested in #XYZ. I tested this in jupyter notebooks and unit tests and it worked as I think it should. I ended up needing to write special treatment for list/dict comprehension and wouldn't be surprised to see more special cases popping up. we'll see
• The max history and ignore modules can now be configured
• The module ignore thing uses regular expressions. This may be overkill and I'm not entirely sure if this poses a security issue. Implementation actually felt even simpler compared to a "split / str compare". I figured that people might want to use glob patterns or something more sophisticated than hard coded paths. regex was the "simplest" choice. If that faces resistance, I'll go for a simple str compare (startswith, is in, etc.)
• I added a poor mans HTML repr (don't be too critical, I just copied stuff around) simply to get started. I hope that somebody with a better sense for style will iterate on this
• I chose to go for a dedicated UUID nevertheless (see Add initial draft of large scale computation class #4972 (comment)) mostly because I believe we need a UUID, especially when connecting to an external service and I couldn't find the UUID @mrocklin was referring to.