Skill versioning & update strategy for `diffusers-cli skills`

Following up on the discussion around the `diffusers-cli skills` command (added in #13966), we need to figure out a longer-term strategy for versioning and updating skills.

Current state:
- `diffusers-cli skills` uses the same skill-management semantics as the `hf` CLI (`add` / `list` / `preview` / `update`).
- Any skill under `.ai/skills` can be installed via the CLI.
- Skill updates are handled via `diffusers-cli skills update `.
- Installing a skill pinned to a specific version is **not** supported yet.

Open questions / things to decide:
- **Versioning & coupling.** How (if at all) should skills be coupled to `diffusers` package versions? Pinning skills to releases vs. keeping them decoupled so they can be updated without a package release.
- **Hosting.** Should we host skill files in a bucket (like the `hf` CLI does at `huggingface.co/buckets/huggingface/skills`) rather than bundling them in the package? This would let us update skills independently of releases.
- **CI / publishing workflow.** Consider a workflow (similar to the community-pipeline mirroring workflow) that tests skills and then pushes them to the bucket.
- **Offline install.** Skills are just markdown — do we want to also bundle them for offline installation?
- **Docs.** Document the update/versioning behavior on the [CLI docs page](https://github.com/huggingface/diffusers/blob/main/docs/source/en/using-diffusers/cli.md#skills).

Related: #13966, #13381.

---
Requested by Sayak Paul - [Slack thread](https://huggingface.slack.com/archives/C065E480NN9/p1783711247069459?thread_ts=1783711247.069459&cid=C065E480NN9) - [Agent trace](https://huggingface.co/buckets/huggingface/moon-bot-memory/tree/sessions/2026-07-24T10-27-40-971Z_5ede90f1-1b0b-408d-88d4-c24e5492742d.jsonl)