We removed the ability to automatically load extensions from jars
many years ago, but this test remained. It only continued to pass
because some other library loaded the 'json' library which meant
that the tested classes were available. When this same example is
run outside of the test suite, it fails:

```
$ cx jruby-10.0.2.0 ruby -e "require 'uri:file:./lib/ruby/stdlib/json/ext/parser'; p JSON::Ext::Parser"
NameError: uninitialized constant JSON::Ext::Parser
  const_missing at org/jruby/RubyModule.java:4866
         <main> at -e:1
```

Because this functionality no longer exists, I am removing the
test.

A few additional stdlib were added in 4.0 so this archive is now
a bit larger.

* Removed SortedSet and related files (gem TBD).
* Removed stdlib .rb stub.
* Fixed inspect output, intersect? with Array, and infinite
  enumerables for new tests.
* Moved @hash variable to a bound Java field.

* Fix inspect output expectations.
* Guard SortedSet tests behind a require, since that collection is
  now in a separate library (knu/sorted_set). These tests probably
  should move to the external library once we work out the details
  (see knu/sorted_set#7)

Based on code from CRuby by Tomoya Ishida:

ruby/ruby#13680

When the incoming block's signature is important to the iteration
method being called, the previous logic would hide that signature.
The change here propagates it through the CallBlock.

This does not exactly match the new native Set impl, but it passes
specs and tests related to initialize.

This updates some logic related to the new native Set.

Only fails are now related to us not tracking iteration-in-progress
for either Set or Hash.

Mostly surrounding the handling of `$/`, `$\`, and `$,` with
regards to duping and freezing and assignment of non-nil warnings.

CRuby has for a long time had separate C globals for the default
record separator (`rb_default_rs`) and the configurable record
separator (`rb_rs`), while we only used a single global slot for
both. This led to failures expecting that the configurable version
could be configured using the `-0` flag (as in `-072`) without
affecting e.g. the output of `puts`.

This change may modify behavior in JRuby where we go after the real
global variable (rather than the JRuby global default separator)
but as the `-0` flag and the `$/` and `$-0` globals are being
deprecated, I'm not really worried about it.

This was updated in 2015 but not tested until recently.

See ruby/ruby@f830ace

Now that to_set does not force a size, only endless is rejected.

See ruby/ruby#15173

* All vars if nil is given
* Error if non-nil, non-array given

When I switched to using fetch, I did not notice it also coerces.
Remove the extra coercion here.

I always forget which methods return null and which return UNDEF.
We should unify that some day but for now at least document it.

CRuby has both rb_num2dbl and rb_num_to_dbl, which have slightly
different logic and raise slightly different errors. This patch
matches that to allow both forms of errors to be thrown in the
appropriate places.

CRuby frequently checks if a given VALUE is "immediate", which
in their terms means it has no presence on the heap. We do not
have immediates in JRuby, but we mark these types as such for
optimizations that check for it.

* False is not immediate but True is, so lower that override.
* Float is both in CRuby, but we mark it immediate for fast-path
  logic that checks this.

We are using this to represent all core singleton literals (nil,
false, true), fixnum-ranged integers and all floats and symbols.
This restores immediate status to false.

A Data object should be frozen even if it has no members

This makes the spec pass on Ruby 4.0.0 and earlies as well. This saves
some issues with the next spec sync.

The style `.should.predicate?` is slightly more idiomatic, and seems to
be preferred (see ruby/mspec#80 for a related
discussion).

…thods

Remove Process::Status#& and Process::Status#>>

Add ruby_bug for spec of Data with no members being frozen

Fix Data subclasses with ivars