feat(selection): add advance_on_view to LabelingSelection

[RapidPoseidon]

Why

A/B experiments measure accuracy by position (1st, 2nd, 3rd session per person). With Shuffled/Sequential, a task an annotator saw but skipped is served to them again, which skews that measurement. The backend now has LabelingSelection.advanceOnView (rapidata-backend#5592, live in prod) to move the annotator past a task once it has been shown. This exposes it in the SDK as LabelingSelection(..., advance_on_view=True).

Builds on the regenerated client from #941 (merged).

Flow

flowchart LR
  user["LabelingSelection(advance_on_view=True)"]:::changed ==>|"_to_model()"| model["ISelectionLabelingSelection.advanceOnView"]
  model ==>|"POST /job (selections)"| order["Order / Campaign service"]
  classDef changed fill:#fff3bf,stroke:#f08c00,color:#000

Loading

API	Change	Before → After
LabelingSelection.__init__	changed	— → advance_on_view: bool = False; ValueError with Random

Decisions

• Random with advance_on_view=True raises at construction time, so callers get the error up front rather than as a backend 400 on job creation.

Verified

New tests/rapidata_client/selection/test_labeling_selection.py; pyright src/rapidata/rapidata_client reports 0 errors.

Agent skill unchanged: src/rapidata/_skill/ does not document LabelingSelection or retrieval modes.

🔗 Session: node-04de9e35

🤖 Generated with Claude Code

[github-actions]

⚠ Agent skill not updated

This PR does not modify src/rapidata/_skill/. That directory is the agent skill that ships in every SDK release and that coding agents read through python -m rapidata skill.

Before merging, pick one:

1. The change affects what an agent needs to know (new or renamed API, changed parameter, default or result field, new gotcha): update src/rapidata/_skill/SKILL.md (or a companion guide next to it) in this PR.
2. Nothing the skill documents changed: a reviewer applies the skill-unchanged-approved label. New commits remove the label again.

[RapidPoseidon]

The only red check is Agent Skill, and that's the intended review gate, not a code failure: build, test and type-check are green. src/rapidata/_skill/ (SKILL.md, reference.md, examples.md, flows-for-preference-data.md) never mentions LabelingSelection, retrieval modes or max_iterations, so adding advance_on_view changes nothing the skill documents. A reviewer needs to apply skill-unchanged-approved to clear it.

[RapidPoseidon]

After the rebase onto main (ccc033c, now that #941 is merged), build, test and type-check are green. Agent Skill is still the only red check. It is the reviewer gate, not a code failure: src/rapidata/_skill/ never mentions LabelingSelection or retrieval modes, so applying skill-unchanged-approved clears it.

[RapidPoseidon]

@LinoGiger thanks for the approval! Agent Skill is still red because the skill-unchanged-approved label is missing (the skill doesn't document LabelingSelection). Could you add it? I'll merge as soon as the check turns green.