feat: add lowest and lowest-direct resolution strategies

[Victor-Su-Ortiz]

Summary

Adds --resolution-strategy with highest (the existing default), lowest, and lowest-direct, using the normal poetry.lock and environment workflow. Related to #3527.

lowest prefers low compatible versions throughout the graph. lowest-direct prefers low root dependencies while keeping the existing preference for transitive dependencies. These are preferences subject to the complete constraints and existing prerelease/yanked behavior, not a guarantee that every package reaches an independent minimum.

Existing locks remain authoritative unless --regenerate is supplied. The option is also exposed for install/sync when a lock must be created. Invalid values are rejected. Includes CLI documentation and offline tests for strategy selection, constraints/backtracking, extras, Python compatibility, prereleases, lock retention/regeneration and deterministic locking.

Validation

• Independent related solver/provider/lock/install/installer suite: 816 passed, 2 expected xfails.
• Related model-run suite including sync: 896 passed, 1 skipped, 2 expected xfails.
• Configured pre-commit hooks passed for all changed files.
• ruff format --check src tests, ruff check src tests, and mypy src tests passed.
• Independent actual CLI integration used a local wheel index to generate and install all three strategies, verified selected installed versions, repeated byte-identical locks, and a constraint that overrides the preferred low version.
• Full suite: 3318 passed, 33 skipped, 2 expected xfails, 1 setup error. The self-update fixture FileExistsError also reproduces on an unchanged baseline; unrelated self-update code was left intact.

This is an AI-assisted implementation with independent review and testing. Submitted as a draft for review of the CLI/API design and compatibility; it has not been merged or released.

Pull Request Check List

Related: #3527

• Added tests for changed code.
• Updated documentation for changed code.