docs: rename the "set" concept to "stream" and document jm.set deprecation

[konn]

Description

Renames the "set" concept to stream throughout the tutorials, and documents the deprecation of jm.set in favour of jm.stream.

The tutorials used to introduce the concept as a "set" and then walk it back with an admonition explaining that JijModeling "sets" are really streams — ordered, lazily evaluated, and allowing duplicates. That two-step explanation is dropped: the concept is simply called a stream from the start, and a shorter note now states plainly that a stream is not a mathematical set (pointing at jm.unique for deduplication).

Changes, in both en and ja:

• basics/expressions — the section becomes "Streams and comprehensions for sum/product"; the "sets are streams" admonition is dropped entirely (with the concept simply called a stream there is nothing left to disambiguate — the duplicates-allowed fact and the jm.unique pointer are folded into the surrounding prose); the array-iteration and dictionary-iteration admonitions are reworded away from "behaves like a set"; explicit conversion is now jm.stream. A new note records the 2.8.0 rename and that jm.set survives as a deprecated alias.
• basics/modeling — stream-vs-scalar comparisons; "streams of indices" instead of "index sets".
• references/migration_guide_to_jijmodeling2 — Element is replaced by first-class streams rather than Sets; jm.set(expr) → jm.stream(expr), with a note about the deprecated alias.
• releases/unreleased — a new Feature Enhancements entry explaining why the rename happened, showing jm.stream, and demonstrating that the deprecated jm.set still works, including its Decorator API comprehension form.

Historical release notes (releases/jijmodeling-2.5.0, 2.5.1) still mention jm.set; they describe what shipped at those versions and are left as-is. Wording that means a mathematical set — "key set", "edge set", "set of labels" — is untouched.

Re-executed notebooks

The rename changes rendered model output — expressions print as stream(..) instead of set(..), types as Stream[..] instead of Set[..] — so every notebook that displays a model was showing stale output, including ones whose prose I did not touch (introduction, quickstart/*, advanced/*, references/cheat_sheet, …).

They are re-executed here against a local build of the JijModeling branch that introduces jm.stream, using task sync_paired_notebooks plus the two releases/unreleased notebooks (excluded from the book build by _config.yml, so executed explicitly). The unreleased notes now carry real output for both the jm.stream example and the deprecated-jm.set example.

Notebooks whose only diff was execution metadata (iopub timestamps, kernel patch version) are deliberately left untouched, so the diff shows only output that genuinely changed.

task check_paired_notebooks and textlint both pass for the two languages.

⚠️ The RTD preview still shows the OLD printer output — this is expected

Do not read the ReadTheDocs preview as the result of this PR. Pages such as basics/placeholders still render

Placeholders: {'A': Placeholder(name="A", dict, dict_keys=set((N, L)): ...

even though the committed notebook says dict_keys=stream((N, L)).

The reason is that the preview does not use the committed outputs at all:

• docs/{en,ja}/.readthedocs.yaml installs the tutorials package with pip install .[dev], so jijmodeling>=2.0.0,<3 resolves to the latest released version (2.7.1) — a clean install that knows nothing about this branch's jm.stream.
• docs/{en,ja}/_config.yml sets execute_notebooks: force, so the build discards every committed output and re-runs the notebooks against that released library.

The only pages that render committed outputs are the ones _config.yml excludes from execution — releases/jijmodeling-* and releases/unreleased*. That is why releases/unreleased does preview correctly, showing jm.stream(N) and the deprecated-jm.set example.

Nothing is pinned or worked around for this: no pre-release of JijModeling is published just to make the preview green. The preview corrects itself as soon as 2.8.0 reaches PyPI, since the existing dependency range already admits it.

Suggested order: merge the JijModeling PR → release 2.8.0 → rebuild this preview → merge to develop.

CI is likewise unaffected: check_book_* re-executes from scratch rather than diffing against the committed outputs, and check_paired_notebooks does not execute at all.

🤖 Generated with Claude Code

[NY57]

LGTM!

[NY57]

LGTM!