Prepare for v0.6 release

[fsschneider]

This PR prepares for the upcoming v0.6.0 release of the benchmark. Depends on #874.

Specifically, it changes:

• Rework Readme to be clearer and easier to understand.
• Remove deprecated COMPETITON_RULES and CALL_FOR_SUBMISSION docs.
• Backfill the changelog with the most relevant changes and highlight it in more documents.
• Increment rules version to 0.6.0 and detail new rule changes:
  • Reduce studies from 5 to 3.
  • Remove held-out workloads.
  • Update API in documentation.
• Integrate and describe the new version policy (Proposal: Unify Versioning for Codebase, Leaderboard, and Rules #870).
• Minor formatting improvements.

Todos

There are a few more rule changes to detail. I also want to give the entire DOCUMENTATION.md a thorough review, focusing on readability and removing deprecated statements. The changes include, for example:

• Remove mentions of model track?
• Rewrite section on qualification set -> Phrase it as a recommended protocol for reporting partial results if limited resources are available. An AlgoPerf light version?
• Update note on "how many submissions"
• Update FAQs (for rolling leaderboard format)
• Integrated important requirements from the COMPETITON_RULES doc (e.g., Apache 2.0 license, defining software dependencies, CLA/membership/trademark agreement?)
• Mention that you can change the dropout value in the forward pass now.

I would also like to:

• Re-organize the algorithms into a single directory. I have a first draft, but I also need to:
  • Fix the tests, that are currently failing due to reference_algorithms being moved.
  • Add a section in README.md about Algorithms (or Sample Submissions) with a short description of what they are.
  • Proper README.md within the algorithms/ directory to enumerate all subfolders and algorithms.
  • What to do with the prize-qualification-baseline? There is no longer such a thing, but we do want to keep "a baseline", right?
  • Clean-up README.md in the individual algorithms/ subfolders.

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

[priyakasimbeg]

EDIT: Disregard below comment, I just noticed they are in the "algorithms" folder

~ Hi Frank, I like that you reorganized the submissions but I was wondering if the algoperf folder is the right place for them. There are a lot of logs in those folders that will inflate the size of the algoperf package ~