Integrate native Rime and stabilize input, output, and rendering

- Add optional modifier-only Super binding whose tap is cancelled by
  other input, focus changes, grabs, locks, and session teardown.
- Ship the native ferese-rime user service, Rime defaults, installer
  inventory, package dependencies, session environment, and docs.
- Vendor Smithay 8a27023 with the SHM texture-cache format fix and add
  GPU regressions for format changes and resize snapshots.
- Release reused CRTC owners before staging output modes, then restore
  changed outputs before former owners when a transaction fails.
- Wait for resize-capture GPU completion, expand filtered surface
  damage, and preserve IME popup origin under fractional scaling.
- Mark weekends like holidays in both calendar layouts.

vendor/smithay/AI.md

@@ -0,0 +1,30 @@
+The following policy applies if you make a contribution to Smithay's
+code, documentation, issues, comments, artwork or any other type of resource.
+
+It is your responsibility to ensure that:
+- You fully understand how the code in your contribution works.
+- Your change is accurately described in the commit message.
+- Your contribution is tested and working as described.
+- To the best of your knowledge, you are not introducing bugs; any exceptions to this rule must be presented with clear reasoning.
+- You are not violating copyright (such as software licenses, creative commons or any other). (See also our [DCO](https://github.com/Smithay/smithay/blob/master/DCO.md).)
+- Use of any AI generated code or documentation is clearly indicated in your commit messages.
+
+If you repeatedly fail to meet these standards, your contributions will no longer
+be considered by Smithay's maintainers and possibly removed without further notice.
+
+Due to the requirements listed above, we strongly advise against using large
+language models or other probabilistic tools such as Copilot and ChatGPT.
+Code produced by these systems is usually more difficult to reason about
+and fails to express important technical design decisions in a concise form.
+This results in additional maintainer burden, draining the (already limited)
+time and energy of our reviewers. Additionally, the legal circumstances of the training sets these
+tools are based on are subject of active legal proceedings, which puts the copyright of generated
+contributions into question.
+
+Last but not least, we ask you to not use generative AI to write comments or descriptions
+on our issue tracker, discussions and pull requests. In our experience these tools tend to generate
+very inconcise text and more often incorrect statements about the cause of underlying issues or code.
+
+If you rely on such tools to make up for a language barrier, we strongly encourage you to use
+(ML-based or otherwise) translation tools as our maintainers much prefer to deal with
+imperfect english than generated walls of text. Many of us aren't native english speakers as well.