Minor fixes
- Fixed: some smaller/simpler models formatted their response slightly differently (e.g. markdown bold instead of the exact marker) and the summary showed raw, jumbled text instead of the structured result — the parser is now more tolerant of formatting variations.
- Fixed: opening the popup on a PDF showed a confusing "couldn't extract content" error — now shows a clear "PDFs aren't supported yet" message instead.
- Clarified: results depend on how well the chosen local model follows instructions — very small/simple models may occasionally show malformed or inaccurate content. This is a model limitation, not an extension bug.

Fixed translation issues: summaries could get stuck showing the original language after turning the extension on for a site, ignoring the default language set in Options. Also fixed language codes (e.g. "es", "de") sometimes being ignored by the model, which made translation to Spanish and German unreliable.