refactor(scheduler): add type hints to CosineLRScheduler

[haru-256]

Description

This PR adds explicit type hints to the __init__ arguments of CosineLRScheduler.

Motivation and Context

Currently, the warmup_lr_init argument defaults to 0 (integer). As a result, static analysis tools like Pylance (based on Pyright) and linters like Ruff infer the type as int.

When a user passes a float value (e.g., 1e-6) to warmup_lr_init, it triggers a type mismatch error/warning, even though the code handles floats correctly at runtime.

Example error in Pylance:

"Argument of type 'float' cannot be assigned to parameter 'warmup_lr_init' of type 'int'"

Changes

• Imported Tuple, List, and Union from typing.
• Added explicit type annotations to __init__ arguments in timm/scheduler/cosine_lr.py.
• Specifically, warmup_lr_init is now typed as float (defaulting to 0), which resolves the linting error.

Notes

I noticed similar issues in other schedulers, but I am addressing CosineLRScheduler in this PR as a first step.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.