Fix some offsets computation when using scroll

[DreydenGys]

While I was working on a personal project using both goblin and scroll, I found some bugs in the way offset are computed for both DataDirectories and OptionalHeader.
Here some minor fixes without breaking changes.

While going through the code I notice that most parsing methods are present twices:

• one implementation with the scroll traits
• one using the method parse of each structs

Is there a reason why theses two implementations cohabit? Using only one could reduce bugs and improve maintainability.

[m4b]

I’d like to see a test that fixes whatever bug you had, (which fails today without this patch), otherwise I’m not sure exactly what this fixes for you, can you go into more detail? This appears to only change the pwrite impls which won’t affect the parser iiuc?

[m4b]

I’m not sure bad magic is the correct error here, I’m sure there’s something more applicable, but thank you for getting rid of this panic, that’s bad?

[DreydenGys]

I used the same as the one used in src/pe/optional_headers.rs:597 and src/pe/optional_headers.rs:576 to be consistent.

[DreydenGys]

I noticed that for now, the only reason we rely on the method DataDirectories::parse is that we take the count parameters taken from the optional header. The thing is that no matter what we will construct an array of 16 (NUM_DATA_DIRECTORIES) internally. However as we don't store this count number internally, we will write the 16 entries when using TryIntoCtx.
We should either store this count if we know that PE files can have less than 16 entries or consider that it will always be 16 and transform the DataDirectories::Parse implementation a TryFromCtx without the count parameter.