[FE] Display structured transcription JSON in a readable format on the verification screen

[mshkkkkk]

Description
Structured transcriptions are currently displayed as unformatted text on the verification screen, which makes them difficult to read and verify.
As agreed with the Product Owner:

• In view mode, structured JSON should be displayed as structured JSON in a nested tree format
• Edit mode should continue using a regular textarea.
• Edited JSON should be validated before saving.

Acceptance Criteria

• - Valid JSON transcription is displayed in a structured, nested tree format in view mode.
• - Nested objects and arrays are visually distinguishable.
• - In edit mode, the transcription is displayed as text in a regular textarea.
• - The transcription is validated as JSON when the user saves the changes.

[RomanNabukhotniidev]

Before starting, please sync with the team on this one. I heard someone suggested a library for rendering the JSON tree rather than writing our own.

Worth finding out which one it was and using it, if it holds up - a tree view with nesting, collapsing and theming is more work than it looks, and we already have enough on release 3.

[illiaboiko]

Update — implementation reworked from the original ticket description

After syncing with the team, we didn't build a collapsible JSON tree view. Instead, the model's output schema now includes a new page_text field (per-preset, backend-driven) that returns the whole page as continuous, human-readable text — Markdown tables for tabular content, inline (?) / [?] / [...] markers for uncertain, illegible, and missing text — alongside the existing structured records. The verification screen renders that as real paragraphs and tables instead of a JSON dump, with the uncertainty markers shown as highlighted, tooltipped spans. Edit mode is still a plain textarea (per the PO-agreed AC), now sized to fit content and capped so it never pushes the Save/Cancel row off-screen.

Scope note: this ticket only covers displaying and editing the transcribed text readably — it does not cover the rest of the verification workflow (e.g. feeding corrected words back into the lexicon/context for later pages). That's tracked separately:

• [BE] Keep structured output in sync with corrected text #140 / PR TSA-140: [BE] Keep structured output in sync with corrected text #326 — re-deriving structured (and the lexicon) from a human correction
• [FE] Mark lexicon words inline in the transcription text #412 — surfacing lexicon words inline in the transcription text
• [FE] Show the document lexicon and let a word be marked wrong #273 — showing the document lexicon and marking a word wrong

Not yet done from the original AC: JSON validation on save. That criterion doesn't map cleanly anymore now that the edited value is readable text rather than raw JSON — needs a decision on what "valid" means for this format before it's closed out.

PR incoming on branch feat/TSA-331-readable-transcription.