v0.2.1

Fixed

• PDFs whose text layer is legacy non-Unicode "glyph soup" (e.g. some older
  Burmese printing fonts encode their text as Private Use Area glyph ids) no
  longer come back blank. Such a layer is detected as unusable and the page is
  OCR'd instead of returning meaningless characters.
• An expired or revoked Google sign-in is now surfaced clearly. The token is
  validated once before a run, so it fails with a single "sign in again" message
  instead of every page failing with a cryptic error; is_authenticated no
  longer reports an unusable token as signed in.

Added

• When OCR needs Google sign-in, the desktop app now offers to sign in in place
  and retries the run on success.