Account for BSA asset override as sorting criterion

[loot]

Account for BSA asset override as sorting criterion

[Ortham]

I'm finally picking this up, as the changes to adding tie-break edges for loot/loot#1370 change the way that ties may be broken, and this shouldn't be a problem, except that LOOT is blind to the impact that might have on assets loaded through BSAs that get pulled into the game by plugins, so sorting cannot be stable with respect to that.

Adding support for taking BSA conflicts into account won't necessarily avoid the tie-breaking causing changes, but at least those conflicts will be taken into account.

The bsa-overlap branch currently contains a spike adding support for parsing a subset of BSA files and using the results when adding overlap edges during sorting. BSA files that use big-endian numbers aren't supported, and neither are BA2 files - I don't think I've got any BSAs that use big-endian numbers, so I've got nothing to test that against, and I haven't looked into the BA2 file format yet but plan to add support for it next.

There also aren't many tests yet, and those that do exist use official game BSAs I've got installed, so the tests will fail in CI. I've manually verified sorting completes without errors logged for Skyrim SE, but haven't checked the impact of the new sorting logic.

[Freso]

If it can be made to work for BA2s too, that would be great.

[Ortham]

I've added support for BA2 archives and a bunch of tests. I still need to add tests for the sorting changes.

Unfortunately I'm seeing unexpected duplicates across several of Fallout 4's BA2 archives: because BSAs/BA2s include hash values I'm checking for overlap using them rather than path strings, but BA2s use 32-bit hashes vs. BSA's 64-bit hashes so have a much greater chance of collision. However, I'm not sure if the duplicates I'm seeing are due to multiple BA2s containing the same file path or different file paths with hash collisions, I need to do more debugging...

[Ortham]

I switched to calculating 64-bit hashes for BA2 archives and those duplicates were still logged, so I think they must be due to multiple BA2s containing the same file path. I'll stick with the 64-bit hashes though, as some plugins load enough assets that a collision is more likely than not when using a 32-bit hash.

I've also finished adding tests, so this is basically done - I'm only holding off merging because it potentially changes the outcome of LOOT's load order sorting, so I'm treating it as a breaking change. The changes now live in the new-tie-break-logic branch until they're merged.

[Ortham]

Done as of bdd5f8d.