attachments preview overlay, docx/xlsx/odf rendering, composer drag-drop & undo-send UX

Frontend:
- Full-screen attachment preview overlay (AttachmentPreview) for image/pdf/txt
  plus office docs: .docx via docx-preview (rendered live into the sandboxed
  iframe so header/footer VML images survive), .xlsx via xlsx-preview, and
  .odt/.ods/.odp/.odg via a JSZip+DOMParser ODF walker (officeConvert).
- Type-aware attachment icons (green spreadsheet, blue document, red PDF,
  image/archive); new document/image/spreadsheet/print SVGs.
- Composer: drag-drop attachments (body drop inserts images inline, else
  attaches); inserted images are selectable/resizable/deletable (RichTextEditor).
- Undo-send: "Sending…" banner with live "N seconds to Undo" countdown.
- Search result list: pin/archive/delete update correctly; grey-out only for
  search hits; scoped "Search in Trash/Spam".

Backend:
- Search: prefix (search-as-you-type) matching, email-address search, and
  address-first result ordering; Trash/Junk excluded with scoped folder search.
- Attachments: RFC 6266 Content-Disposition for non-ASCII filenames; one-time
  inline image loading.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Rcj3TpydMTzTSmU7kunzf8

backend/app/api/messages.py

@@ -7,6 +7,7 @@ import io
 import uuid
 import zipfile
 from typing import Annotated
+from urllib.parse import quote
 
 from fastapi import APIRouter, Depends, Query
 from fastapi.responses import HTMLResponse, Response
@@ -146,6 +147,15 @@ async def message_cid(
     return Response(content=data, media_type=media_type)
 
 
+def _content_disposition(disposition: str, filename: str) -> str:
+    """RFC 6266 Content-Disposition safe for non-ASCII names. HTTP header values
+    are Latin-1, so a raw Polish/CJK filename (ł, ń, …) crashes the response with
+    a 500; emit an ASCII fallback plus a UTF-8 ``filename*`` for modern clients."""
+    ascii_name = filename.encode("ascii", "ignore").decode("ascii").replace('"', "").strip()
+    fallback = ascii_name or "attachment"
+    return f"{disposition}; filename=\"{fallback}\"; filename*=UTF-8''{quote(filename)}"
+
+
 async def _real_attachments(db: AsyncSession, message_id: uuid.UUID) -> list[Attachment]:
     """Downloadable (non-inline) attachments for a message, in a stable order."""
     rows = await db.scalars(
@@ -234,5 +244,5 @@ async def message_attachment(
     return Response(
         content=data,
         media_type=media_type,
-        headers={"Content-Disposition": f'{disposition}; filename="{att.filename}"'},
+        headers={"Content-Disposition": _content_disposition(disposition, att.filename)},
     )

backend/app/search/pg.py

@@ -257,6 +257,17 @@ class PgSearchBackend(SearchBackend):
         for cond in conditions:
             stmt = stmt.where(cond)
 
+        # Exclude Trash/Junk from search by default (Gmail-style), so deleting a
+        # message from search results removes it. An explicit `in:trash`/`in:junk`
+        # overrides this to search there on purpose.
+        if not _targets_removal_folder(groups):
+            stmt = stmt.where(
+                or_(
+                    Mailbox.special_use.is_(None),
+                    Mailbox.special_use.notin_(_REMOVAL_FOLDERS),
+                )
+            )
+
         decoded = _decode_cursor(cursor)
         if decoded is not None:
             cur_hit, cur_date, cur_id = decoded
@@ -291,6 +302,23 @@ class PgSearchBackend(SearchBackend):
         return hits, next_cursor
 
 
+_REMOVAL_FOLDERS = ("trash", "junk")
+
+
+def _targets_removal_folder(groups: list[list[dict[str, object]]]) -> bool:
+    """True if the query explicitly scopes to Trash/Junk via an ``in:`` filter, in
+    which case the default removal-folder exclusion must not apply."""
+    for group in groups:
+        for clause in group:
+            if (
+                clause.get("type") == "filter"
+                and clause.get("key") == "in"
+                and str(clause.get("value", "")).strip().lower() in _REMOVAL_FOLDERS
+            ):
+                return True
+    return False
+
+
 def _positive_text_terms(groups: list[list[dict[str, object]]]) -> list[str]:
     terms: list[str] = []
     for group in groups: