FEX doesn't handle executable stack flags correctly.

[Sonicadvance1]

Kernel documentation:

   *                 CPU: | lacks NX*  | has NX, ia32     | has NX, x86_64 |
   * ELF:                 |            |                  |                |
   * ---------------------|------------|------------------|----------------|
   * missing PT_GNU_STACK | exec-all   | exec-all         | exec-none      |
   * PT_GNU_STACK == RWX  | exec-stack | exec-stack       | exec-stack     |
   * PT_GNU_STACK == RW   | exec-none  | exec-none        | exec-none      |

FEX implementation:

   *                 CPU: | has NX, ia32     | has NX, x86_64 |
   * ELF:                 |                  |                |
   * ---------------------|------------------|----------------|
   * missing PT_GNU_STACK | exec-none        | exec-none      |
   * PT_GNU_STACK == RWX  | exec-all         | exec-all       |
   * PT_GNU_STACK == RW   | exec-none        | exec-none      |

Effectively if PT_GNU_STACK exists with RWX then we should only enable executable stack.
If PT_GNU_STACK /doesn't/ exist then we should enable exec-all on only x86, but leave it alone on x86-64.

Needs some unittests to ensure correct behaviours for the matrix of six behaviours (We don't emulate a system without NX these days).

Low prio as this isn't affecting anything we know of.

There's also the additional quirk that if an executable is mapped from a noexec filesystem that mappings should inherit that. Not sure if there is a clean way to detect that.... Might only matter with execve, not mmap?