fix: remove deprecated VQModel/VQEncoderOutput shim past target version 0.31

[lorenzozanee]

What does this PR do?

Fixes #14365
Adds a regression test (5 assertions) verifying the shim is gone, the old import
path raises ImportError, and VQModel remains accessible from the public API.

Before submitting

• Did you use an AI agent (Claude Code, Codex, Cursor, etc.) to help with this PR? If so:
  • Did you read the Coding with AI agents guide?
  • Did you run the self-review skill on the diff?
  • Did you share the final self-review notes in the PR description or a comment?
• Did you read the contributor guideline?
• Did you read our philosophy doc? (important for complex PRs)
• Was this discussed/approved via a GitHub issue or the forum? Please add a link to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag members/contributors who may be interested in your PR.

[lorenzozanee]

AI Self-Review: READY

Validated against: .ai/review-rules.md, .ai/AGENTS.md, .ai/testing.md

Checks Run

Check	Result
Regression tests (5/5)	PASS
ruff check	PASS
ruff format	PASS
utils/check_copies.py	PASS
utils/check_dummies.py	PASS
utils/check_support_list.py	PASS
utils/check_forward_call_docstrings.py	PASS
make deps_table_check_updated	PASS
git diff --check	PASS
tests/models/autoencoders/test_models_vq.py	PASS (17 passed, 0 failures)

Upstream-only CI

Jobs that require the upstream PR environment and cannot be verified locally:

• pr_tests.yml (fast tests)
• pr_dependency_test.yml
• pr_torch_dependency_test.yml
• pr_labeler.yml (label automation)

No Blocking Findings

• Coding style: conforms to project conventions
• Copied code rules: not applicable (no # Copied from blocks touched)
• Dead code analysis: the deleted file was unreachable via __init__.py exports; all remaining import paths use the canonical autoencoders.vq_model location
• No debug logs, large files, or sensitive data in the diff