Simplify ET replay and add Integration testing

[briancoutinho]

Summary

Today there are two replay logic supported - one for only comms and one for compute+comm. This has several drawbacks

• Code exists in two directories - compute and comms
• Changes in one side are not testing compatibility with the other.
• @TaekyungHeo has observed bugs and crashes with the replay when both compute and comms is enabled.

Crashes

@TaekyungHeo to add more info on how to reproduce issues

Code unification

Basic idea is to pull things out to a replay directory and unify the code
Details TBD

Integration testing

Ensure changes are unit tested to avoid impact to external users.

[TaekyungHeo]

I am sharing my version of PARAM, which you can find here: https://github.com/TaekyungHeo/param/commits/comms-bugfix/

The most critical issue in commsTraceReplay involves the execution trace parsing logic. A quick fix is available in my branch, although I am uncertain about its correctness.
The second critical issue concerns the construction of execution trace paths. It functions correctly within Meta, but fails externally. I plan to submit a PR to address this.
The third critical issue is in running comms replay and comp replay simultaneously. During my time at Meta, this was not supported, and I am currently unsure if it has been addressed. This issue primarily stems from inconsistent trace path construction logic between comms replay and comp replay.

Additionally, while I was at Meta, I validated the computation replay results. Unfortunately, I have not had the opportunity to validate communication replay or the combination of communication and computation replay.

[TaekyungHeo]

I have created two pull requests (PRs) related to comms trace replay as shown below. Please review the PRs when you get a chance.

• Fix incorrect path construction in commsTraceReplayBench
• Support for SLURM in commsTraceReplay

Additionally, I have identified another issue in parsing execution traces, but I am not sure whether it is a universal bug or not. You can find the bugfix here: TaekyungHeo@f8042dd

There is also another pull request for trace_link.py available at: #90