feat: batch OCR processing for all manuscript pages

[nikazzio]

Summary

Add a "Run OCR on all pages" action that processes every page of a manuscript in sequence (or parallel batches), with a real-time progress bar.

Motivation

Today OCR runs one page at a time via HTMX polling. For a 200-page manuscript this means 200 manual clicks. Batch processing is the single biggest UX improvement for transcription workflows.

Proposed approach

• Add a POST /studio/{manifest_id}/ocr/batch endpoint that queues all un-OCR'd pages as a single job.
• Reuse the existing JobManager / export job infrastructure for progress tracking.
• Stream progress via SSE or HTMX polling (page N/M, estimated time remaining).
• Allow cancellation mid-batch.
• Write each page's OCR result to the transcription DB as it completes (incremental save).

Acceptance criteria

• "Run OCR on all pages" button in Studio toolbar
• Progress indicator (pages done / total, ETA)
• Cancellation support
• Results saved incrementally — partial OCR is preserved on cancel
• No regression on single-page OCR flow

Technical notes

• Rate-limit OCR API calls to respect provider TOS (configurable concurrency).
• Consider a max_concurrent_ocr setting in config.
• Existing src/universal_iiif_core/services/export/service.py job pattern is a good template.