rgblink scramble could be even more scrambled to expose more mistakes

[zlago]

while the scramble flag does serve its purpose of showing when you forget to bankswitch, it introduces two new issues:

1. it hides alignment mistakes, as sections generally get placed at $4000
2. it hides mistakes where data is read out of bounds, since all data past a section is usually $ff

possible solutions for each issue:

1. try to put sections at the end of a bank, or avoid putting sections at $4000 (the former is probably better)
2. allow for randomising section banks instead of putting each section in its own bank

the second one probably isnt as much of an issue as the first, as its me not testing the non-scrambled build..

[aaaaaa123456789]

I'd say that a better fix would be to randomize all ROM padding in scramble mode, and to possibly shift sections by a few bytes when there's free space to do so. However, shifting sections would probably require rerunning the linker...

[evie-calico]

That’s the issue I see with this. No matter how it’s implemented, the misalignment would cause fragmentation, which might cause a scrambled build to fail when a normal one wouldn’t.

[ISSOtm]

The way I see it, allocating sections primarily towards the end of ROMX would help catch OOB accesses there, as then those accesses would tend to fall in VRAM (then it's a die roll as to whether a VRAM access exception is triggered, but better than nothing) or SRAM (which is most often locked).

Additionally, I see no harm in doing thins by default outside of -x mode. (Sure, that would change section order, but this should not be as breaking a change since #136's resolution.)

[evie-calico]

Putting the sections at the end is a really good idea, I didn't consider that.

[zlago]

i didnt intend on solving both problems.. i only really thought of it as "should be a cheap way to put a section at somewhere other than the start"

[Rangi42]

An initial attempt to pack sections from the end of banks instead of the beginning makes it seem harder than one might expect. And unlike some feature requests (like "put two sections in the same bank but I don't care which"...), this one isn't really limiting anything -- scramble mode already works, it just isn't maximally inconvenient for revealing implicit errors. :P So, I feel comfortable tagging this for 2.0 (aka "wait til the Rust rewrite"), although someone can always feel free to pick it up sooner.