Add recognition JSON output and UI viewer

[amirgholipour]

What changed

• Persist the correlation result as a stable JSON recognition artifact in S3.
• Add a Recognition JSON tab to the current UI.
• List recognition sessions and load multi-page JSON results through server-side APIs.
• Provide copy and download actions without exposing S3 access to the browser.
• Recover a usable JSON artifact when a model response contains surrounding text or malformed XML.

Why this change is proposed

The processing pipeline produces specialist and correlated artifacts, but the UI does not provide a direct machine-readable result view. Users can see chat progress yet have no clear place to inspect, copy, or download the final recognition output. Strict parsing can also leave no consumable artifact when otherwise useful model output has minor formatting defects.

This change gives every correlated page a predictable JSON representation and makes those files accessible through the existing authenticated server boundary.

Benefits

• Enables downstream automation without scraping chat text.
• Makes processing results visible and testable in the current UI.
• Supports multi-page sessions while retaining per-page provenance.
• Keeps bucket names, object access, and AWS credentials out of the browser.
• Produces an explicit parse status when recovery was required.

Validation

• Processed a deployed image workflow and verified a correlation JSON artifact was written.
• Verified the Recognition JSON tab can list, load, copy, and download results.
• Verified recovery output remains valid JSON.
• Verified production UI build with pnpm and Python module compilation.
• Reviewed the branch against the current upstream main.

Authored by Saeed Kasmani (amirgholipour).