Dataset info and starter pipelines in CSV File Scan property pane

## Feature Summary

When a user adds a CSV File Scan operator and picks a dataset file, the Property pane shows only file metadata (path, encoding, limit). The user has no quick way to see what's in the dataset, and no guidance on what to add next — which is exactly the moment when help is most valuable.

This proposes two small additive sections inside the CSV File Scan Property pane:

**Part A — File Info**: a read-only panel showing the selected file's metadata (dataset, version, size, uploader, upload time, description).

**Part B — Starter Pipelines**: a default-collapsed callout that suggests a small operator chain (classification / regression / exploration) based on the dataset's column types. One Apply button adds the whole chain to the canvas as a single undo step. No LLM. No auto-execution.

Both sections are purely additive — the user can ignore them, nothing happens unless they click Apply.

This builds on my hackathon prototype #5122, which was too aggressive (it generated and ran entire workflows automatically). This proposal keeps the underlying insight — the moment a user picks a dataset is when help is most valuable — but narrows it to suggestion-only, scoped to one operator's pane.

Related to #4085, #5099, #5394.

## Proposed Solution

I built a working prototype to make this concrete.

**Part A — File Info panel**

Read-only metadata about the selected file, pulled from the existing `DatasetService`. No new backend.

**Part B — Starter Pipelines** (default collapsed)

For Iris, the suggestion is a classification pipeline based on a categorical target (`Species`) and 4 numeric features. The card shows the operator chain, a one-line "Why?", and an Apply button.

Click Apply → all operators are added to the canvas, fully linked, as a single undo step:

## Key design choices (vs #5122)

- Suggestion-only — never auto-runs.
- Deterministic, no LLM.
- Operator ports wired via operator metadata, not hardcoded.
- Single undo step.

## Suggested PR breakdown

1. Part A (File Info) — frontend-only, low risk.
2. Part B (Starter Pipelines + Apply) — after Part A merges.

Would appreciate feedback on whether this direction makes sense.

## Affected Area

Workflow UI

Start with the CSV File Scan Property pane and the existing DatasetService, then review prototype #5122 and the related issues for context. Done means the pane has additive file metadata and a collapsed, deterministic starter-pipeline suggestion with an Apply action that adds one linked chain as a single undo step without auto-execution.

Written by the indexing model from the issue text.