Implement a more reasonable CPU fetching flow

[bigspider]

While #20 added a basic cache for pages, the way memory is accessed during execution allows for much deeper optimizations:

• except after branches, the next instruction is almost always in the same page
• code pages are immutable

Moreover, currently the CPU fetches, decodes and executes a single instruction. However, code pages are immutable, so there is no reason to decode an instruction multiple times (unless the page was dropped).

Improving this flow will bring a huge performance boost.