Fix opcode table over-read crash on Windows

[tbdye]

Problem

machine68k.CPU() crashes with ACCESS_VIOLATION on Windows. The segfault occurs during m68k_init() → m68ki_build_opcode_table(), before any user code runs.

Root cause

The opcode handler table is terminated by a zero sentinel {0, 0, 0, {0,0,0,0,0}}. The first while loop in m68ki_build_opcode_table() scans forward looking for mask == 0xff00 as a phase transition:

while(ostruct->mask != 0xff00)

However, m68kmake never generates entries with mask 0xff00. The loop processes all ~1967 valid entries, reaches the zero terminator, evaluates 0 != 0xff00 as true, and continues reading past the end of the array.

On Linux/macOS this happens to work — the over-read lands in zero-filled BSS, and the subsequent == loops (mask == 0xf1f8, mask == 0xfff0, etc.) all exit immediately on mask=0, so the function completes despite the undefined behavior. On Windows, the .rdata section layout places the const array near a page boundary, and the over-read crosses into unmapped memory.

The subsequent loops are safe because they use == checks, which naturally terminate when they encounter a different or zero mask value. Only the first loop, which scans forward with !=, has this problem.

Fix

Add a zero-terminator guard to the first loop so it stops at the end of the table:

while(ostruct->mask != 0xff00 && ostruct->mask != 0)

Verification

• CPU(1) no longer segfaults on Windows (tested Python 3.13 and 3.14)
• Machine(1, 1024) works correctly
• 41 tests pass, 34 skipped (pre-existing [remote]/[greenlet] skip conditions)