fix(ai): use documented Claude 4.6 model IDs

[lukisch]

The AI assistant sends claude-sonnet-4-6-20250514 by default and claude-opus-4-6-20250514 when Opus is selected. These identifiers mix the 4.6 generation with an older snapshot date. Anthropic documents the canonical IDs as claude-sonnet-4-6 and claude-opus-4-6.

Correct those two enum values. Keep Haiku 4.5 and the surrounding request behavior unchanged. Add three fixture-only regressions that exercise AIService.complete and inspect the actual SDK model/message payload for the default, Opus and Haiku selections. Record the correction under Unreleased; no product version change.

Primary reference: https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions

Validation on Windows Python 3.12 / pytest 9.1.1:

• Unchanged base: 257 tests passed.
• New payload proof before repair: Sonnet and Opus failed on the invalid suffix; Haiku passed.
• After repair: complete suite 260 passed; full Ruff and git diff --check passed.
• No real provider/model request or credential-store access in the new tests. Native CI still requires terminal readback; this is not live Anthropic service acceptance.

Author model: gpt-6.1-sol (Codex). Draft pending independent review; merges disabled. Based directly on master 49d7a83, independent of the action-upgrade PR #8. Canonical worktree and OneDrive deployment stay on their existing states until reviewed integration.

[github-actions]

Welcome! 👋 Thanks for your first pull request in this repository.

A maintainer will review it soon. Please make sure:

• Your changes are tested
• Documentation is updated if needed
• The PR description explains what and why

Thanks for contributing!