[Hackathon] feat: add smart source and visual trace

[tanishqgandhi1908]

Video Submission

https://youtu.be/KlyswVrWLZU

What changes were proposed in this PR?

This PR improves the end-to-end data experience for hackathon workflows by making ingestion smarter, image workflows first-class, and visual outputs easier to understand.

Motivation

Before:

User task	Current friction
Load a dataset	Users must choose the right source operator before they know the file format
Read a folder	There is no source input that can read a folder and process all files in it together
Work with images	There is no first-class way to load image datasets as structured input
Understand a visual result	Users can see the final output, but not how it was produced

After:

User task	New experience
Load a dataset	Smart Source auto-detects file type, dialect, and schema
Read a folder	The same source can read a folder of similar files together and preserve source-file lineage
Work with images	Image folders become structured input rows with real image previews
Understand a visual result	Clicking a visual result can open a Visual Journey side panel

Main changes

1. Add Smart Source (SmartFileScan) with support for CSV, TSV, JSON, JSONL, Arrow, Parquet, Excel, images, and plain text.
2. Add backend file inference plus frontend inference summaries so the property panel can show detected format, delimiter, header status, sheet, schema size, and folder counts.
3. Extend folder support across dataset selection and file scanning:
   • folders can be selected from the dataset picker
   • FileScan can read folders while preserving relative file names
   • new File Split operator routes rows from the same source file to the same output port
4. Make image workflows more natural:
   • image folders produce rows containing image bytes plus format and dimensions
   • recognized image binaries are serialized as image data URLs
   • result tables render image thumbnails instead of raw binary text
5. Teach the agent service about SmartFileScan and include operator display names in the prompt so the agent can reason about user-facing operator names such as Smart Source.
6. Add a reusable Visual Journey side panel:
   • visualizers can emit rich trace payloads
   • ordinary image clicks fall back to a structural upstream workflow trace
   • iframe-origin clicks are handled correctly so visualizer interactions open the side panel reliably

Any related issues, documentation, discussions?

• Related to hackathon discussion Task ideas for the dkNet-AI · Apache Texera Agent Hackathon #5059.

How was this PR tested?

PATH="/Users/tanishqgandhi/.bun/bin:$PATH" bun test agent-service/src/agent/prompts.test.ts agent-service/src/types/agent.test.ts

JAVA_HOME=$(/usr/libexec/java_home -v 17) sbt "testOnly org.apache.texera.amber.operator.source.scan.smart.CSVDialectSnifferSpec org.apache.texera.amber.operator.source.scan.smart.FormatDetectorSpec org.apache.texera.amber.operator.source.scan.smart.SmartFileSourceOpDescSpec org.apache.texera.amber.operator.source.scan.smart.SmartFileSourceOpExecSpec org.apache.texera.amber.operator.fileSplit.FileSplitOpDescSpec org.apache.texera.amber.operator.fileSplit.FileSplitOpExecSpec org.apache.texera.amber.operator.source.scan.file.FileScanSourceOpDescSpec org.apache.texera.web.service.ExecutionResultServiceSpec"

PATH="/Users/tanishqgandhi/.nvm/versions/node/v24.15.0/bin:$PATH" yarn ng test --watch=false --include='src/app/workspace/service/visual-trace/visual-trace.utils.spec.ts' --include='src/app/workspace/component/visual-trace-panel/visual-trace-panel.component.spec.ts' --include='src/app/workspace/component/result-panel/result-table-frame/result-table-cell.utils.spec.ts'

Manual verification:

1. Loaded folder-backed CSV datasets through Smart Source.
2. Loaded an image folder and confirmed result cells render image thumbnails.
3. Opened an HTML visualizer, clicked a winner card, and confirmed the Visual Journey panel opens from iframe-origin clicks.