I recently wrote klondike in Rust after getting frustrated with bugs in my friend's LLM-written version. I highly recommend it as a recreational programming exercise, the first 4 hour blitz filling out the types was a blast. Later, trying to make as many invalid moves unexpressable by the given types as possible was a fun challenge. I ended up with a 232 byte struct for the board state with all values stack-allocated. The only way to make it considerably smaller would probably be to encode card permutations.
It's notable that on most CPUs, the comparison instruction is equivalent to a subtraction but without writing the actual result.