Support 'bulk' interfaces

[kosarev]

We currently only support handlers like on_step() and on_read()/on_write() that mandate the emulators to execute a single instruction or provide a mean to access a single memory cell. This prevents us from attempting to do some kinds of optimisations, e.g.:

• memmove()-like implementations for LDIR/LDDR (means better performance, mentioned in The emulator is not as fast as it's advertised. :-P #16 (comment)).
• Fast HALT/HLT emulation.
• Complete the support for NMOS/CMOS Z80 differences #35 requires knowing it in advance how much code we can execute without being interrupted.
• Just-in-time compilation for hot basic blocks. Better performance again.
• Support port handlers that deal with blocks of input/output data rather than with individual bytes. It would particularly be useful for the Python API to void calling handlers on every IN/OUT instruction (then https://github.com/kosarev/zx can be updated to use that).