The basic idea here is to avoid pushing a frame for block_given?
when it is known to be our core implementation. Instead, we use
indy to test the block the current method was called with and
return a result directly. When the block_given? is not our
core impl, it falls back on a normal indy invocation. This allows
us to remove the frame requirement from block_given? callers, and
the cost of the block_given? call largely disappears.

Caveats:

* This is very much hacked in place, with duplicated code in
  multiple places.
* In order to pass through the block we were given, we add a Block
  parameter to the call site. If someone defines a block_given?
  in Ruby, it would receive our given block.