Use lru_cache to suppress memory usage of commutation analysis

[t-imamichi]

Summary

Use lru_cache to suppress memory usage of commutation analysis. It also simplifies the code a little.
This might be oversimplification because it does not take care of qargs.

Details and comments

[jakelishman]

Do you have memory profiles for this change? The current cache only stores a single boolean per key, so I'm surprised it has too large a memory footprint as it is.

I'm also very surprised that DAGOpNode is hashable in the general case. This seems a little dangerous; they store mutable values within themselves, and they allow themselves to be mutated, so by the normal Python rules, they shouldn't be hashable.

[t-imamichi]

Hi, @jakelishman. Thank you for your comment. @ewinston has the memory profile data with large number of qubits.
I saw #6982 after I made this PR. My PR oversimplified the code, so it is not likely to run. I will close it.