Nice work! As it stands, you seem to be doing pure interpretation for the CPU; have you considered dynamic recompilation? You can see this in practice for the 6502 here: https://github.com/daeken/Dyna6502Asm

In essence, you accumulate instructions in basic blocks and turn them into functions, which are then 'executed' (meaning turned into function objects) and then cached. In your case it becomes a little more complex as you do cycle timing, interrupts, etc, but there's no reason you can't bake this into your basic block functions. This drastically increases the performance of CPU emulation, while maintaining the same high quality.

If you're interested in doing this and/or have any questions, feel free to contact me.