[perf] refactor: optimizes DataProtoFuture to use fractional lazy fetching of futures

[yurun00]

What does this PR do?

This PR optimizes the DataProtoFuture by replacing the brittle collect_fn and dispatch_fn mechanisms with robust, native chunking logic associating futures with start_fraction and end_fraction.

Key changes:

• Removed obsolete collect_fn and dispatch_fn properties from DataProtoFuture.
• Added properties start_fraction and end_fraction to track chunking offsets in start and end futures directly.
• Implemented fractional range calculations in chunk to accurately track boundaries across multiple chunks.
• Streamlined the get method to materialize data efficiently using integer offsets, avoiding overhead of fetching unnecessary futures
• Updated tests/test_protocol_on_cpu.py to cover the new fractional splitting logic.

Co-authored-by: Antigravity

Checklist Before Starting

• Search for similar PRs. Paste at least one query link here: is:pr is:open DataProtoFuture
• Format the PR title as [{modules}] {type}: {description} (This will be checked by the CI)
  • {modules} include single_controller
  • {type} is refactor

Test

Ran the updated CPU unit tests locally to validate the new fractional splitting logic:

pytest tests/test_protocol_on_cpu.py

cpu unit test: https://github.com/yurun00/verl/actions/runs/25240216728
sgl: https://github.com/yurun00/verl/actions/runs/25240216710
vllm: https://github.com/yurun00/verl/actions/runs/25240216727

API and Usage Example

No API change

Design & Code Changes

As shown in key changes above

Checklist Before Submitting

Important

Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review.

• Read the Contribute Guide.
• Apply pre-commit checks: pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always
• Add / Update the documentation.
• Add unit or end-to-end test(s) to the CI workflow to cover all the code. If not feasible, explain why: ...
• Once your PR is ready for CI, send a message in the ci-request channel in the verl Slack workspace. (If not accessible, please try the Feishu group (飞书群).)
• If your PR is related to the recipe submodule, please also update the reference to the submodule commit via git submodule update --remote or cd recipe && git pull origin main.

[CLAassistant]

All committers have signed the CLA.

[gemini-code-assist]

Code Review

This pull request refactors the DataProtoFuture class in verl/protocol.py to use fractional offsets (start_fraction and end_fraction) for data chunking, replacing the previous collect_fn and dispatch_fn mechanism. The changes include a more precise chunk implementation using the fractions module and an updated get method that slices data based on these fractions. Additionally, the PR includes minor cleanups to np.reshape calls and introduces comprehensive unit tests for DataProtoFuture in tests/test_protocol_on_cpu.py. A potential precision issue was identified in the get method where floating-point multiplication followed by integer truncation could lead to off-by-one errors in indexing.