Fixes #3799

When we set `default_proc` to nil, we also have to update the internal
flag.

Fixes
#4302

Fix marshaling Hash with default_proc set to nil

I had introduced a bad optimization during IR build phase.  It assumed
all Strings could be built out of order but I did not account for the
fact that the same StrNode could be referenced by two instructions at
the same time (since each occurrence will perform a dup).

Note: It might look like I removed FrozenStrings from building out of
order but the ordinary logic in buildWithOrder will end up creating
the operand in place and not generating the copy since it is an
ImmutableLiteral.

Fix IO#syswrite. Fixes #3799

Add missing PortUnreachable catch. Fixes #4104

With #4321 merged, we no longer have
to skip that ruby spec.

Skip less specs

https://bugs.ruby-lang.org/issues/12574

The basic fix is that in respond_to? we do not return true for protected
methods.  I made a few internal changes:

1. deprecated is RubyModule.isMethodBound(name, checkVisibility, checkRespondTo)
2. added RubyModule doesMethodRespondTo(name, checkVisibility)
3. changed what I think is appropriate consumers to call doesMethodRespondTo
   but in worst case it will call isMethodBound(name, checkVisibility) which
   is what it would have done anyways before this change.  The only risk is
   that I converted something which wants isMethodBound but that would indicate
   some method further upstream calling a respond-to method and not really
   wanting respond to behavior.  The things I changed we obvious in that they
   had respondto names.
4. I made a Helpers.doesMethodRespondTo so that RespondToCallsite and
   RubyModule can share the same logic.