confuses portmaster, which (understandably) expects that we didn't
fake up our bootstrap dependency.
Instead, only set up a BUILD_DEPENDS for a bootstrap if there isn't one
installed. A more complicated and arguably more correct patch would
always set up the BUILD_DEPENDS for the bootstrap JDK but would do so
correctly based on what it found to use.
This almost certainly needs to happen to the other jdk* ports, but I'm
going to await some positive feedback before doing so.