Follow up on public api review notes

[333fred]

• Project.RemoveDocuments: Everything else in this type uses IEnumerable<T>, not ImmutableArray<T>. If we could start again from day 1 we might make everything that, but given that everything today uses the former, we should do so here. This applies to the overloads on Solution as well.
• Solution.RemoveAdditonalDocuments/RemoveAnalyzerConfigDocuments: We added these to Solution, but not to project, even though that also has the invididual Remove methods. We should add them to Project as well.

[jasonmalinowski]

Project.RemoveDocuments: Everything else in this type uses IEnumerable, not ImmutableArray. If we could start again from day 1 we might make everything that, but given that everything today uses the former, we should do so here. This applies to the overloads on Solution as well.

So @333fred and @CyrusNajmabadi, it looks like we're already a bit inconsistent here. Solution.AddDocuments (which is already a shipped API) is already taking an ImmutableArray, so when we added RemoveDocuments we were being consistent with that. I can switch the Remove methods to IEnumerable, but then that just results in a different inconsistency. Is that something we feel is necessary here?

[jasonmalinowski]

Also as a reminder any change to a signature here induces a fairly high test cost given we need to take any chance here through VS shiproom.

[333fred]

I think I'd prefer consistency with the symmetric method (AddDocuments) than with the rest of the methods in Solution (which may or may not be related). It would have been good if they were all ImmutableArray to start with, but we are where we are.

[jasonmalinowski]

@333fred: so in that case, the only item to do then is adding the Project.Remove* methods that match the Solution.Remove* methods? OK if we do that in 16.7 so as to avoid QB process for this?

Tagging @tmat too on the IEnumerable vs. ImmutableArray since he was messing with some of that recently too.

[333fred]

I think that's fine.

[jasonmalinowski]

@tmat or @CyrusNajmabadi any concern about fixing just the missing methods, i.e. the second bullet point and not the first?

[jasonmalinowski]

Given it's now too late to fix anyways, going with the PR that only fixes the second issue.