The input parser's CSI and SS3 state handlers do not check for ESC
(0x1B). Per ECMA-48 §5.3.1, ESC is always effective and should restart
the escape sequence machine from any intermediate state. Instead, ESC
falls into the "bad parse" branch in inpStateCsi and is silently
discarded, and in inpStateSs3 it fails the key lookup and is lost.

This causes the byte(s) following the discarded ESC to be
misinterpreted. In practice, the user's first keystroke after
Suspend/Resume can be swallowed: engage() sends escape sequences that
may provoke terminal responses, and if those responses leave the parser
mid-CSI, the next ESC-prefixed key (arrow keys, function keys) is
consumed as the tail of the stale sequence.

Add an ESC check to both inpStateCsi and inpStateSs3 that transitions
to inpStateEsc with the standard timeout setup, matching the behavior
of inpStateInit.