Harden Smoke Signal noise learning and JSON decision writes

Improve Smoke Signal noise-pattern learning and matching

feat: multi-line noise patterns — paste one per line, all saved on reject

fix: detected font display, remove reviewer, add noise paste box for NON_STORY_TEXT blocking

feat: persistent noise pattern exclusion — reject NON_STORY_TEXT to block on next OCR run

refactor: remove profile tab, fix nan font, full OCR progress bar, renumber tabs

refactor: strip ingest to bare minimum — PDF + story pages + one button

fix: ocr unpack 3 vals, no duplicate rename, detected font in review, Reviewer label

feat: confidence override checkbox — reviewer can mark perfect pages as 100%

feat: single-button pipeline — Register + OCR + Review in one click

fix: ocr_done state triggers review load after OCR completes; add Gruffalo word corrections

feat: prev button, step instructions all tabs, noise threshold 3.2, word corrections I->T G->s

feat: add step instructions to ingest, greyscale always, OCR progress indicator

fix: yield after each Surya batch — live progress, no more silent hang

fix: use token length as primary noise discriminator — filters td/wht/LN illustration noise

fix: tighten illustration noise filter — alpha ratio, per-region cleanup

fix: restore save_book_scope + merge OCR push into single handler

fix: remove stale _save_scope_and_push_ocr causing UnboundLocalError

fix: scope->OCR wiring after ocr_page_selection defined, no more UnboundLocalError

fix: merge scope save into single click — push story pages to OCR fields

fix: move scope->OCR wiring after ocr_page_selection defined (UnboundLocalError)

feat: auto-populate book ID, safe title, and OCR page scope from ingest fields

fix: convert run_ocr to generator — live log streaming, no more silent hang

fix: remove dangling except block causing SyntaxError

fix: font_name reset per page, phase logs, per-book error surfacing

restore: revert smoke_signal_tab.py to 0964303 (undo bad overwrite)

fix: font_name reset per page, phase logs, per-book error surfacing

Fix profile dtype crash by stringifying manifest page_count

Fix ingest upload path resolution to prevent empty manifest

Make rights/page-scope updates resilient and auto-resolve latest book

Fix save-page scope dtype crash and prefill OCR selectors from ingest scope

Switch ingest IDs to code+year and handle duplicate ID rename

Fix runtime crash on unsupported show_copy_button

Add persistent book page scope, safe titles, and OCR stall guards

Add manual OCR page exclusion by PDF page number

Add manual OCR page selection and deterministic review ordering

Integrate punctuation corrector into OCR routing and review learning

Embed supplied banner art as text data URI and refine hero

Remove duplicate OCR progress bars and redesign Smoke Signal hero banner

Reduce tesseract hard batch timeout default for faster failover

Prevent indefinite OCR hangs with hard tesseract batch timeout

Show minimal OCR activity indicator without full progress bar

Hide floating OCR progress overlay

Sync wizard underline with selected pipeline tab

Auto-populate review item on tab open and after OCR run

Gate fallback OCR by minimum confidence to reduce gibberish queue

Consolidate profile into ingest and make profile tab optional

Tune fallback for speed and filter low-confidence OCR noise

Accelerate Tesseract fallback with parallel OCR and timeouts

Speed up fallback OCR rendering and keep progress moving