feat: inline member management on Marmot group info screen

- Fold the Remove Member screen into the info screen: each member row
  now has an inline remove icon that opens a confirmation dialog.
- Fold the Add Member screen into the info screen: a search field with
  a user-suggestion list lives directly above the member list, so
  adding a member never requires leaving the screen.
- Move Leave Group out of the scrolling body into a top-bar action so
  it is reachable regardless of member count.
- Shrink the relay strip (35dp tiles with a small activity dot) and
  tuck it next to the group header; drop the "Relays" label and the
  MLS epoch readout, which were visual clutter.
- Route the chat screen's Add Member button to the info screen and
  delete the two standalone screens and their routes.
- Add headless interop test 17: A creates a group, adds B, removes B,
  re-adds B, and verifies B can both receive and send messages. Guards
  the reported regression where a re-added member came back without a
  usable leaf and silently lost the ability to post.

https://claude.ai/code/session_01JaeqZwVNLKvUUvXWRzPmRP

tools/marmot-interop/headless/tests-manage.sh

@@ -155,6 +155,95 @@ test_08_admin_promote_demote() {
   fi
 }
 
+test_17_readd_after_remove() {
+  banner "Test 17 — Re-add after remove restores B's leaf"
+  local id="17 re-add after remove"
+
+  # Scenario reported on the info screen: A creates a group, adds B,
+  # removes B, then adds B again. Before the fix B rejoined without a
+  # usable own_leaf, so B could see metadata but could neither decrypt
+  # new posts nor sign messages. This test catches that regression by
+  # driving a full round-trip (A→B and B→A) after the re-add.
+
+  local out gid mls_gid
+  out=$(amy_json marmot group create --name "Interop-17-readd") || {
+    record_result "$id" fail "amy create failed"; return
+  }
+  gid=$(printf '%s' "$out" | jq -r '.group_id')
+  mls_gid=$(printf '%s' "$out" | jq -r '.mls_group_id')
+  [[ -n "$gid" && -n "$mls_gid" ]] || {
+    record_result "$id" fail "missing ids from create"; return
+  }
+
+  # A adds B.
+  amy_json marmot group add "$gid" "$B_NPUB" >/dev/null || {
+    record_result "$id" fail "initial add B failed"; return
+  }
+  wait_for_invite B 60 >/dev/null || {
+    record_result "$id" fail "B never received first invite"; return
+  }
+  wn_b groups accept "$mls_gid" >/dev/null 2>&1 || true
+
+  # Confirm the baseline works: A → B and B → A both deliver.
+  amy_json marmot message send "$gid" "pre-remove hello" >/dev/null || {
+    record_result "$id" fail "A send before remove failed"; return
+  }
+  wait_for_message B "$mls_gid" "pre-remove hello" 90 || {
+    record_result "$id" fail "B missed pre-remove hello"; return
+  }
+  wn_b messages send "$mls_gid" "pre-remove pong" >/dev/null 2>&1 || true
+  amy_json marmot await message "$gid" --match "pre-remove pong" --timeout 60 >/dev/null || {
+    record_result "$id" fail "A missed B's pre-remove pong"; return
+  }
+
+  # A removes B.
+  amy_json marmot group remove "$gid" "$B_NPUB" >/dev/null || {
+    record_result "$id" fail "remove B failed"; return
+  }
+  # Let the Remove commit propagate to B's wnd.
+  local deadline=$(( $(date +%s) + 120 )) removed=0
+  while [[ $(date +%s) -lt $deadline ]]; do
+    if ! wn_b --json groups members "$mls_gid" 2>/dev/null \
+         | jq -e --arg p "$B_HEX" '(.result // .) | .[]? | select((.pubkey // .public_key) == $p)' \
+         >/dev/null 2>&1; then
+      removed=1; break
+    fi
+    sleep 3
+  done
+  [[ "$removed" -eq 1 ]] || warn "B still appears in members after remove — continuing"
+
+  # B republishes its KP so the next add can find a fresh one.
+  wn_b keys publish >/dev/null 2>&1 || true
+  sleep 3
+
+  # A re-adds B.
+  amy_json marmot group add "$gid" "$B_NPUB" >/dev/null || {
+    record_result "$id" fail "re-add B failed"; return
+  }
+  wait_for_invite B 60 >/dev/null || {
+    record_result "$id" fail "B never received second invite"; return
+  }
+  wn_b groups accept "$mls_gid" >/dev/null 2>&1 || true
+
+  # Regression probes: after re-add B must have a usable leaf again.
+  # (1) A→B: proves B can decrypt new group messages.
+  amy_json marmot message send "$gid" "post-readd hello" >/dev/null || {
+    record_result "$id" fail "A send after re-add failed"; return
+  }
+  wait_for_message B "$mls_gid" "post-readd hello" 120 || {
+    record_result "$id" fail "B didn't decrypt post-readd message (missing own_leaf?)"; return
+  }
+  # (2) B→A: proves B can still sign/commit — the exact breakage reported.
+  wn_b messages send "$mls_gid" "post-readd pong" >/dev/null 2>&1 || {
+    record_result "$id" fail "B send after re-add failed (likely no leaf)"; return
+  }
+  if amy_json marmot await message "$gid" --match "post-readd pong" --timeout 120 >/dev/null; then
+    record_result "$id" pass
+  else
+    record_result "$id" fail "A never saw B's post-readd pong (B couldn't type)"
+  fi
+}
+
 test_11_leave_group() {
   banner "Test 11 — Leave group"
   local id="11 leave group"

tools/marmot-interop/marmot-interop-headless.sh

@@ -123,3 +123,4 @@ test_13_keypackage_rotation
 test_14_wn_removes_a
 test_15_wn_member_leaves
 test_16_wn_keypackage_rotation
+test_17_readd_after_remove