THUMB: Implement `Load/Store Register offset` instruction

[AlessioC31]

• Add write_word function to internal memory
• Implement instruction
• Rewrote get_bits in order to avoid using for and using binary arithmetics instead. It's faster than 50% on u32 (but this 50% is only 20ns, lol).

Should we move ReadWriteKind somewhere else since it's not used only by single_data_transfer?