I think clang format was simply not designed for this. Rules are not things it applies incrementally, the program is instead built around, parsing your entire program and forgetting (most of) the old whitespace, and generating new whitespace based on the rules you select.
You can see some overview of the architecture here: http://www.llvm.org/devmtg/2013-04/jasper-slides.pdf
First it runs clang lexer and parser, then it divides groups of tokens into “unwrapped lines” which are “tokens we would like to put together on a single line if there was no column limit”. Then a layouter determines the formatting of each unwrapped line based on the various constraints and optimizing for the various penalties.
So, I don’t think “one clang-format action” is actually a thing, the design looks pretty much monolithic to me.