feat(r2): C5 the lightbox, and the in-place client

- On a gallery booth the marks panel moves into a sticky verdict aside
  beside the set. The aside comes first in the document, so a narrow screen
  stacks the question above the work; grid areas place it on the right when
  wide. Nothing in an ordered collection moves. Boards are unchanged.
- The flag tray lists flagged items by tile number: the declared change
  from the panel list's (created, id). The standalone marks page keeps the
  list.
- Inline group headers are divs, never figure.item.
- Every mark-dependent element is a data-region: the verdict, each tile,
  the rail's filter counts. There is also a server-rendered status line.
- The in-place script (base.html) POSTs with an explicit JSON Accept, then
  on 204 swaps every region from a fresh GET. Live media and per-viewer view
  state are carried across the swap, so there is no layout jolt and no
  stopped track. It never re-POSTs: on failure it says so and reloads. Tile
  controls re-bind after a swap, and the grid cursor survives it.
- The `n` key opens the tile's closed note disclosure before focusing it.
- test_embed_browser's keyboard-flag test expected a navigation, which is
  the defect R2 removes. It is updated as declared in the contract, and
  tightened: a window marker must survive, proving no reload.

Browser tests: flag in place, with no reload and no scroll jump, and the
tile, tray and rail count all updated; and a failed save that reloads
without re-POSTing. Two mutations turn them red (no carry, no rail region).
700 passed.

tests/test_flow.py

@@ -407,3 +407,72 @@ def test_a_booth_without_images_shows_its_kind_instead(tmp_path):
     _booth(tmp_path, "songs", {"a.mp3": b"ID3", "b.mp3": b"ID3"})
     row = re.search(r'data-booth="songs".*?</article>', _client(tmp_path).get("/").text, re.S).group(0)
     assert "♪ audio" in row and "<img" not in row
+
+
+# ---- C5: the lightbox ---------------------------------------------------------
+
+def _region(body: str, rid: str) -> str:
+    m = re.search(r'<(\w+)[^>]*data-region="%s"[^>]*>' % re.escape(rid), body)
+    assert m, f"no region {rid}"
+    tag = m.group(1)
+    # regions in these templates do not nest a same-named tag inside themselves
+    end = body.index(f"</{tag}>", m.end())
+    return body[m.start():end]
+
+
+def test_the_verdict_sits_beside_the_set_on_a_gallery_booth(tmp_path):
+    """The tracer for C5: the open question, the flags and the notes live in
+    one aside next to the grid — not in a panel above it that scrolls away."""
+    from booth.marks import declare_pick
+    b = _booth(tmp_path, "g", {"a.png": PNG, "b.png": PNG})
+    declare_pick(b, "q", {"prompt": "Which one?", "options": ["a", "b"]})
+    body = _client(tmp_path).get("/b/g/").text
+    aside = _region(body, "verdict")
+    assert aside.startswith('<aside class="verdict"')
+    assert 'action="/b/g/answer"' in aside and "Which one?" in aside
+    assert body.count('action="/b/g/answer"') == 1, "the pick renders once, in the aside"
+    assert 'class="lightbox"' in body and 'id="grid"' in body
+
+
+def test_the_flag_tray_lists_flags_by_tile_number_not_by_click_order(tmp_path):
+    """Flag #3 first, then #1: the tray reads #1, #3. Today's panel list is in
+    click order `(created, id)`; the tray is the SET's order, the declared
+    change. Each entry is the original shown small, blurred if the item is."""
+    from booth.app import set_blurred
+    b = _booth(tmp_path, "g", {"a.png": PNG, "b.png": PNG, "c.png": PNG})
+    set_flag(b, "c.png", True)
+    time.sleep(0.01)
+    set_flag(b, "a.png", True)
+    set_blurred(b, "a.png", True)
+    aside = _region(_client(tmp_path).get("/b/g/").text, "verdict")
+    tray = re.findall(r'<a class="tray-item( is-blurred)?" href="view\?f=([^"]+)"[^>]*>.*?#(\d+)<', aside, re.S)
+    assert tray == [(" is-blurred", "a.png", "1"), ("", "c.png", "3")]
+
+
+def test_groups_get_an_inline_header_that_is_not_a_tile(tmp_path):
+    """When the rail finds grouping informative, each group's first tile is
+    preceded by a header — a div, never a figure.item, so the tile sequence
+    the keyboard and the order check walk is exactly the items."""
+    files = {f"{g}-{n}.png": PNG for g in ("aa", "bb", "cc") for n in (1, 2)}
+    _booth(tmp_path, "g", files)
+    body = _client(tmp_path).get("/b/g/").text
+    grid = body[body.index('id="grid"'):]
+    heads = re.findall(r'<div class="grp-head"[^>]*><span class="grp-key">(\w+)</span> <span class="grp-n">(\d+)</span>', grid)
+    assert heads == [("aa", "2"), ("bb", "2"), ("cc", "2")]
+    assert len(re.findall(r'<figure class="item', grid)) == 6
+    # and no header at all when grouping is not informative
+    _booth(tmp_path, "flat", {f"DSC{n}.jpg": PNG for n in range(4)})
+    assert 'class="grp-head"' not in _client(tmp_path).get("/b/flat/").text
+
+
+def test_every_mark_dependent_element_is_a_swappable_region(tmp_path):
+    """The in-place script replaces regions by id; a stale flag count or tile
+    after an in-place flag is the failure this pins."""
+    b = _booth(tmp_path, "g", {"a b.png": PNG, "c.png": PNG})
+    body = _client(tmp_path).get("/b/g/").text
+    for rid in ("verdict", "filters", "item-a%20b.png", "item-c.png", "status"):
+        assert f'data-region="{rid}"' in body, rid
+    # the standing board has no lightbox and therefore no verdict aside
+    _booth(tmp_path, "links", {"links.md": b"- [x](http://x/) <sub>\xc2\xb7 a \xc2\xb7 2026-09-01 10:00</sub>\n"})
+    board = _client(tmp_path).get("/b/links/").text
+    assert 'data-region="verdict"' not in board and 'class="lightbox"' not in board