Migrate the Python stack from Poetry to uv

[khaykingleb]

Modernize the Python project using the conventions from khaykingleb/investment-workbench: PEP 621 metadata, Hatchling, uv dependency groups and locking, a pinned current Python toolchain, Ruff, ty, pydoclint, and pytest.

This migration also establishes a conventional package and CLI identity by renaming the src import package to ml_experiments and exposing an ml-experiments command.

Depends on #12.

Scope

• Replace Poetry metadata and locking with PEP 621, Hatchling, and uv.
• Resolve the newest mutually compatible Python and ML dependency stack.
• Rename Python imports and Hydra _target_ references from src to ml_experiments.
• Add an installed ml-experiments CLI entry point.
• Update local setup, Make recipes, pre-commit, CI, release automation, Docker, and documentation.
• Align quality tooling with the finance repository where it fits this codebase.

Acceptance criteria

• pyproject.toml uses PEP 621 metadata and Hatchling.
• uv.lock replaces poetry.lock, and stale lockfiles fail validation.
• Python 3.14 is used if the complete current PyTorch, TorchAudio, Torchvision, Lightning, and project test stack verifies; otherwise the newest mutually supported Python is pinned and documented.
• Current compatible project dependencies are locked reproducibly.
• Imports and Hydra _target_ paths use ml_experiments.
• The installed ml-experiments command exposes the existing CLI behavior.
• Poetry is absent from project metadata, local setup, hooks, CI, releases, Docker, and documentation.
• Ruff, ty, pydoclint, pytest, and uv lock checks are configured consistently.
• CPU imports/tests and an audio-stack smoke test pass from a frozen uv environment.
• Release versioning has one authoritative source and no Poetry commands.

Out of scope

• Replacing Make with just; tracked separately
• Redesigning the experiment runtime
• Kubernetes submission
• Storage backend integration