In preparation for using the StackWalker API on Java 9+ this patch
streamifies both the native Java stack elements and the frames
from JRuby's interpreter. The former case will allow us to use
the StackWalker API to produce those frames lazily, drastically
reducing the cost of short stack walks on large stacks. The latter
eliminates the copying of the interpreter stack frames and just
streams them directly from ThreadContext.

This improves the performance of limited-depth Kernel#caller by
a substantial amount, dependent on the depth of the stack. For the
following benchmark with a 500-deep stack, the improvement is well
over 10x.

bench_caller_single.rb
```ruby
def foo(depth)
  if depth > 0
    foo(depth-1)
  else
    caller(1, 1)
  end
end

loop {
  t = Time.now
  i = 0
  while i < 10000
    i+=1
    foo(500)
  end
  puts Time.now - t
}
```

Results on 8 and 9:
```
[] ~/projects/jruby $ pickjdk 5 ; jruby blah.rb
New JDK: jdk1.8.0_181.jdk
6.725401
6.5270790000000005
6.555323
^C
[] ~/projects/jruby $ pickjdk 4 ; jruby blah.rb
New JDK: jdk-9.0.4.jdk
0.765762
0.5783969999999999
0.444722
0.442188
0.439658
0.446652
^C
```