There was a very cool project like that just recently. Not exactly the same idea, it uses inlining to specialize a VM definition of a PEG into a parser state machine at comptime. I was impressed.
I have a vague idea of how to write a Zitron template which would do the same basic thing with LALR tables, but I haven’t tried it yet. For one thing I want enough grammars and test data such that I could benchmark them against each other and see if it helps at all.