Detection, done like sync:

- ESC[?u goes to every VT100-like terminal at attach, whatever the
  options.
- A reply turns on a new kittykeys terminal feature. Users can force
  it on or off with terminal-features kittykeys, including kittykeys@
  to turn it off.
- Changing extended-keys now takes effect immediately for attached
  clients (options.c).

Use, per client, when extended-keys is on: Kitty for clients that have
kittykeys, modifyOtherKeys for clients with extkeys, VT10x otherwise.

Pane side:

- Applications can request either protocol; Kitty wins if one asks for
  both.
- Each application gets what it asked for, whichever client typed the
  key.
- ESC[?u is answered truthfully.
- The per-client gate is gone. window.c and popup.c are simply
  master's versions again.

extended-keys always is back to its master meaning, forcing
modifyOtherKeys mode 1 only. Forcing Kitty on every program would have
sent bash ESC[97;5u for Ctrl-A.

extended-keys-format kitty is still accepted, but now behaves the same
as csi-u.

regress/kitty-keys.sh

@@ -13,8 +13,7 @@ TMUX="$TEST_TMUX -LtestK$$ -f$CONF"
 
 trap 'rm -f "$CONF" "$OUT"; $TMUX kill-server 2>/dev/null' 0 1 15
 
-printf '%s\n' 'set -g extended-keys on' \
-    'set -g extended-keys-format kitty' >"$CONF"
+printf '%s\n' 'set -g extended-keys on' >"$CONF"
 
 wait_for_output()
 {
@@ -90,11 +89,13 @@ wait_for_output
 check_output 1b5b3f31751b5b3f3075
 
 : >"$OUT"
+# "always" forces modifyOtherKeys mode 1, not Kitty keys.
 $TMUX set-option -g extended-keys always
 $TMUX respawn-pane -k -t: \
     "stty raw -echo; printf '\033[>0u\033[?u'; dd bs=1 count=5 2>/dev/null | od -An -v -t x1 >'$OUT'; sleep 5"
 wait_for_output
-check_output 1b5b3f3175
+check_output 1b5b3f3075
+wait_for_mode 'Ext 1'
 $TMUX set-option -g extended-keys on
 
 : >"$OUT"