Implementation of query_stats function

[Yejashi]

This PR implements the query_stats function, introducing query language functionality to the statsframe. This enhancement allows end users to perform queries based on data within the statsframe.

Summary of Changes

To achieve this, extensive modifications were necessary. Specifically, all existing stats functions now return a list of output column names. This adjustment was essential to support the caching functionality required for the query_stats function.

Further Explanation

When the query_stats function is called with the squash flag enabled, nodes are sometimes merged into one. This creates two primary issues:

1. A mismatch between the expected number of nodes in the performance data and the statsframe.
2. Invalid data within the statsframe, as some nodes have been merged and their values aggregated.

To address these issues, the statsframe needs to be recalculated after a squash operation. To facilitate this recalculation, I added a caching mechanism. This mechanism records the most recent stats operations and stores them in statsframe_ops_cache. These cached operations are then reapplied on a new statsframe.

By implementing this caching functionality, we ensure that the statsframe remains consistent and accurate after squash operations, enabling reliable query results.

Important Notice

For the methods in this PR to work in the future, we must ensure a standard across all stats function such that they must return a list of output column names. This is an integral aspect of the caching mechanism.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 95.69892% with 4 lines in your changes are missing coverage. Please review.

Project coverage is 80.56%. Comparing base (9c53fd9) to head (d51de55).

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #157      +/-   ##
===========================================
+ Coverage    80.15%   80.56%   +0.40%     
===========================================
  Files           52       53       +1     
  Lines         3442     3535      +93     
===========================================
+ Hits          2759     2848      +89     
- Misses         683      687       +4     

Files	Coverage Δ
thicket/tests/test_query_stats.py	98.48% <98.48%> (ø)
thicket/thicket.py	86.66% <88.88%> (+0.12%)	⬆️

[ilumsden]

There's one major issue right now with query_stats and the corresponding tests:

They don't update or check the statsframe.

Since you are querying the statsframe, I would expect the output Thicket to contain the following:

1. A statsframe containing only nodes captured by the query
2. A graph containing only nodes captured by the query
3. A performance dataframe containing only nodes captured by the query
4. An unmodified metadata table

Your code right now produces Thicket objects with elements 2, 3, and 4, but it doesn't add the filtered statsframe (i.e., element 1) back to the new Thicket.

[ilumsden]

Very good, but there's still a few things that need to be addressed

[ilumsden]

This is very close to being good to go. I've got a few more changes, but I suspect this PR will probably be good after that.

[ilumsden]

I suspect that the next review will be an approval for this PR.

There are just a handful of minor things to change. Most of them amount to removing stringification of multi-level column names for the cache. The remaining changes are just minor one line things.

[ilumsden]

Looks good! @pearce8 this PR is ready for your review

[pearce8]

@Yejashi Do we have docs/examples (tutorial?) to go with this PR?
@Yejashi @ilumsden Any further thoughts on how we could enforce that the future functions use the decorator? Example of using this with a different decorator (e.g., profiling the functions)?

[Yejashi]

@pearce8 , in terms of documentation, there really isn't any meaningful difference on how the stats functions are called. All that's really changed from the user end of things is that the stats functions now return a list of output columns. This is already indicated in the docstrings and but not in the thicket documentation website. A concise sentence could accomplish this task.

In terms of queries, i think we should add a small section in the query portion of the tutorials indicating that the query language can now be used on the statsframe, albeit with a slight difference. An example or two would be enough for this.

[ilumsden]

@pearce8 I mentioned this on Slack in more detail, but this branch shows an example of using Befikir's cache decorator with other decorators: https://github.com/TauferLab/thicket/tree/memray_decorator_example