Add auto version cleaning for aggr and global model version keeping

[ZiyueXu77]

Description

Currently we control the number of model/aggregator versions on server and on clients only with "max_model_history", as model versions scale, this can be both:

• limiting as it can be hard to set the history to a proper value
• wasting resource as some later versions can become obsolete before the earlier versions due to device latency variations

Therefore, updated the behavior of device selection dict to keep record of model versions to reflect the current active model versions, only those active will be kept, and others will be considered obsolete and removed since no device is working on them anymore. This saves resource while keep all updates' contribution without making hard-cut removal. The max_model_history can still be set, with the default updated to inf (apply all updates to global model).

Types of changes

• Non-breaking change (fix or new feature that would not break existing functionality).
• Breaking change (fix or new feature that would cause existing functionality to change).
• New tests added to cover the changes.
• Quick tests passed locally by running ./runtest.sh.
• In-line docstrings updated.
• Documentation updated.

[Copilot]

Pull Request Overview

This PR implements an auto version cleaning feature for managing model and aggregator versions more efficiently. Instead of relying solely on hard-coded history limits, the system now tracks active model versions based on devices currently working on them and only keeps those versions in memory.

• Replaces fixed max_model_history limits with intelligent version tracking based on active device assignments
• Changes default values from finite integers to float("inf") for max_model_history and max_num_active_model_versions
• Adds device wait timeout functionality to handle insufficient device scenarios gracefully

Reviewed Changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
nvflare/edge/updaters/emd.py	Updates aggregator version cleanup logic to use active model versions from device selection
nvflare/edge/tools/edge_recipe.py	Changes ModelManagerConfig defaults to infinity and adds validation for float/int parameters
nvflare/edge/tools/edge_job.py	Updates configure_client method to accept infinity values for max_model_versions
nvflare/edge/executors/edge_model_executor.py	Updates EdgeModelExecutor constructor to support infinity values
nvflare/edge/assessors/model_update.py	Adds device wait timeout functionality and integrates active version tracking
nvflare/edge/assessors/model_manager.py	Adds abstract keep_model_versions method to ModelManager interface
nvflare/edge/assessors/device_manager.py	Adds used_devices tracking to base DeviceManager class
nvflare/edge/assessors/buff_model_manager.py	Implements keep_model_versions method and updates version cleanup logic
nvflare/edge/assessors/buff_device_manager.py	Updates device selection logic and improves device availability checks
examples/advanced/edge/jobs/pt_job.py	Removes hardcoded max_model_history values to use new defaults

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[yanchengnv]

See my comments.

[ZiyueXu77]

/build

[ZiyueXu77]

/build