Release - V0.4.0: Cleanup and launch Semantic retriever.

[MarkDaoust]

Description of the bug:

In #168 we added semantic retriever, but didn't get to resolve all open comments (it's a big feature).

We're merging it without exposing the feature in the public API.

• Resolve the open comments (my last review and @TYMichaelChen's).
• Expose it in the public interface.
• Create a new release.

[MarkDaoust]

AQA -> This should have the same quick access properties as GenerateContentResponse > .parts and .text.

[MarkDaoust]

• Remove retriever.Corpus class in favor of retriever_types.Corpus class. #185
• Fix list methods, add decode functions. #186
• Added create time and update time fields to Corpus, Document, and Chu… #187
• Wrapping glm.RelevantChunk as a class  #190

[MarkDaoust]

#191

[MarkDaoust]

#193

Something's still wrong with the chunk names.
Why are the tests passing full-names instead of short names?

Whatever name the user passes we should combine with the document's full name, right?

[MarkDaoust]

#191

[MarkDaoust]

#197