Performance improvement for Transforms::greedy_pauli_optimisation

[jpacold]

Transforms::greedy_pauli_optimisation makes extensive use of lookup tables defined in GreedyPauliOptimisationLookupTables.hpp. For at least some of these, the use of unordered_map seems to have a significant performance overhead for large circuits.

For example, this function call in the tests takes ~100 seconds on my machine (release build compiled with g++13). Profiling shows roughly $8 \times 10^8$ calls to each of these two functions, which each do one lookup in an unordered_map. Replacing it with an array reduces the time to ~80 seconds.

There might be more room to improve performance with other STL usage adjustments. If this seems worth looking into, I can spend some more time profiling and open a PR.

[yao-cqc]

@jpacold Thanks for profiling this! If replacing unordered_map with an array already gives a ~20% speedup, it sounds like an effective optimisation. Feel free to open a PR with what you find!