refactor(baselines): Updated Floco baseline to Flower 1.26.0

[birnbaum]

Issue

Description

The Floco baseline was still based on Flower 1.18.0 and e.g. not using the new Message API.

Related issues/PRs

#4941

Proposal

Explanation

This PR updates the Floco baseline to Flower 1.26.0 by migrating client/server/strategy code to the current app/message-based API and aligning config/dependencies accordingly.

Main changes:

• Migrated baseline to the current Flower app stack and refactored model utilities to simplify model creation and state handling for Floco under the new API.
• Refreshed baseline plots after re-running all experiments (still about 8h on an M1 Mac).
• Set baselines/floco/run.sh executable
• Updated documentation: Added Floco to top-level README.md.

[Copilot]

Pull request overview

This PR updates the Floco baseline to Flower 1.26.0 by migrating the baseline to the current message-/app-based API (ServerApp/ClientApp, Message, RecordDict, ArrayRecord/MetricRecord) and aligning the baseline’s config, scripts, and documentation accordingly.

Changes:

• Migrated Floco client/server/strategy to the Flower 1.26 app + message-based APIs.
• Updated baseline dependency/config in pyproject.toml and added an experiment runner script.
• Refreshed plots/results tooling and surfaced the baseline in the repo’s top-level README.

Reviewed changes

Copilot reviewed 9 out of 12 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
baselines/floco/run.sh	Adds a convenience script to run all Floco/FedAvg experiments for both splits.
baselines/floco/pyproject.toml	Bumps Flower dependency to >=1.26.0, updates tooling targets, and updates config keys (e.g. fraction-train).
baselines/floco/floco/strategy.py	Ports Floco strategy to flwr.serverapp.strategy.FedAvg and message-based training/eval configuration + aggregation.
baselines/floco/floco/server_app.py	Refactors to ServerApp entrypoint, centralized eval callback, and result serialization.
baselines/floco/floco/client_app.py	Refactors to ClientApp train/evaluate handlers using Message + records; keeps Floco+ personalization via context.state.
baselines/floco/floco/model.py	Refactors model creation/state handling and updates simplex classifier implementation for the new API.
baselines/floco/floco/dataset.py	Adjusts Context import and typing updates for dataloader helpers.
baselines/floco/floco/plot_results.py	Updates results parsing/plotting to match the new JSON output layout and metric keys.
baselines/floco/_static/CIFAR10_Fold.png	Updated plot artifact for the Fold split.
baselines/floco/_static/CIFAR10_Dirichlet.png	Updated plot artifact for the Dirichlet split.
baselines/floco/README.md	Minor formatting tweaks and updated run instructions context.
README.md	Adds Floco to the top-level list of Flower baselines.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

In _SimplexLinear.__init__, _initialize_weight(self.weight) accesses self.weight, but weight is overridden as a property which depends on self._weights (not yet initialized). This will raise an AttributeError during initialization. Consider not overriding weight (override forward instead), or initialize _weights using the underlying Linear parameter via self._parameters["weight"]/a separately stored tensor shape before defining/using the computed weight.

Suggested change

	self._weights = torch.nn.ParameterList(
	[_initialize_weight(self.weight) for _ in range(endpoints)]
	base_weight = self._parameters["weight"]
	self._weights = torch.nn.ParameterList(
	[_initialize_weight(base_weight) for _ in range(endpoints)]

[birnbaum]

Not sure about the two failing checks, the PR title seems to be in the right format and if I run isort locally, it has no complains 🤷🏻‍♂️