Nova 2.1.0

Cleaner text in scraped books, sharper PDF chapter openers, and a system menu that speaks.

• Books from web-scraping sources read cleanly now. Some EPUBs carry deliberately poisoned text: letters swapped for identical-looking foreign characters (so a word sounds fine but search can't match it and character navigation hits gibberish), invisible characters stuffed between letters, and obfuscated site-name watermarks injected mid-paragraph. Nova now restores the real letters, strips the invisible characters, and removes the watermarks. Genuinely foreign-language text — a Russian book, a quoted Greek phrase — is never touched: the cleanup only acts where lookalikes are posing as English.
• PDF chapter openers read in their true order. Key-terms panels laid out in two side-by-side columns used to interleave their entries line by line and occasionally weld two terms onto one line; they now read down each column in turn, one term per line. Opener introduction paragraphs also flow as continuous prose again instead of one printed line at a time.
• PDF contents pages keep their section numbers. On some contents pages the numbers ("4-4", "4-5", ...) separated from their titles and read out later as a bare list; every entry now reads whole — number, title, page — on its own line. The last few index entries that ran together are fixed by the same change.
• Alt+Space works with a screen reader. Opening the system menu used to snap speech straight back to the reading area, and the menu only started reporting once you arrowed. The menu now announces properly when it opens — matching how current Windows apps behave.