Introspection: Reconsider ACTIVEBYTECODE interaction with CODESEPARATOR

We got the following feedback from the author of Spedn:

I'm not quite sure what to think about OP_ACTIVEBYTECODE interaction (or rather the lack of it) with OP_CODESEPARATOR. If the former respected the latter, that could be used to avoid exceeding MAX_SCRIPT_ELEMENT_SIZE in case we're evaluating a large script not wrapped in P2SH. It would also provide a fine distinction between OP_ACTIVEBYTECODE and OP_INPUTINDEX OP_UTXOBYTECODE behavior for that non-P2SH coins with the latter ignoring OP_COCESEPARATOR.