Add ability for PCK patches to remove files

[reduz]

• This was "kindof" existing in PCK but not properly done or supported. It seems this work was initiated over a decade ago and never completed.
• If a file has offset zero in a PCK, it will be considered removed and it will be removed if previously added.

This adds the possibility to add proper directory level diffing to #97118.

[mihe]

At the request of @reduz I've gone ahead and changed this PR to now integrate into the patch creation that was added in #97118. So now if you export a patch and files are found to have been removed since the last export (i.e. with all the patch packs loaded) it will automatically write these new file removal entries to the PCK.

I did have to change things around a bit from how they were originally in this PR though:

1. A removal is no longer signified with a magic zero offset, as it turned out that a zero offset is in fact a valid offset (for the first file) since the offset is relative to the files section of the PCK and not the start of the PCK itself. Now instead we use a new PackFileFlags flag called PACK_FILE_REMOVAL.
2. I removed the previous (incomplete) checking for offset in PackedData::try_open_path mentioned in the PR description. It serves no purpose as PackedData::files will never actually contain any removals with how this feature is done now.
3. Because I'm forced to retrieve the list of previous files from PackedData, which doesn't store the actual paths but instead stores everything in a tree structure, I end up with a list of paths that are not only simplified (as in String::simplify_path) but that also have had any res:// scheme stripped from them. This leads to having to perform this sort of "normalization" elsewhere in order for certain comparisons and hashes to evaluate correctly, so you'll see .simplify_path().trim_prefix("res://") added in a number of places.

One benefit of that last point is that the paths stored in the PCK will from now on never include the res:// scheme, which should slim down PCKs a little bit. As mentioned above these schemes are being stripped when loaded into the PackedData tree structure anyway, so it looks to me like they are currently just bloating the PCK.

[akien-mga]

Looks good to me.

[Repiteo]

Thanks!