Context switch
--------------
        vPC vAC vLR vSP vCPUselect      -> 8 bytes
        sysFn sysArgs0..7               -> 10 bytes
Minimal:
        vPC vAC sysFn vCPUselect        -> 7 bytes

Preemptive context switching is *easier*, because there's no nice
way to swap context from within vCPU itself without leaving anything
behind...

        -> There is instruction to change state
        -> SYS needs sysFn in state
        -> Except through DOKE, needing a special ZP location
        -> CALL and CALLI instuction clobber vLR
        -> Can we special-case an existing instruction? (if stack becomes zero, ...)
        -> Poke vCPUselect (and wait), but then we lose our own state

        From vCPU code:
                LDI vCPUrelease
                ST  vCPUselect
        _spin   BRA _spin
                <continue>

                LDI  SYS_ContextSwitch_40
                STW  sysFn
                SYS  40

        From v6502 code:
                LDIM  v6502release
                STZ   vCPUselect
        _spin   BNE   _spin

        Or better:
                Add a BRK instruction

Interrupt
---------
        Its the same?

Interrupts set: vPC
Interrupts save: vPC vAC sysFn vCPUselect -> 7 bytes

This is almost a full context switch. We can easily swap out and in 18 bytes then?


        ld [y,x]        ; 3 cycles per context in byte -> 18*3 = 54
        st [y,x++]
        st [$dd]

        ld [$dd]        ; 2 cycles per context out byte -> 18*2  = 36
        st [y,x++]

        Total: 90 cycles

        Line 40 -has- 104 cycles