Please take a look at this PR: @nicklan, @zachschuermann, and @scovich (I don't have the rights to add you as reviewers).

It seems like we need some unit tests for this change? Negative tests can be tricky (maybe best avoided), but we should at least have positive tests that the newly allowed types and schemas are actually allowed and behaving as expected?

@scovich I uploaded a completely new version that doesn't modify derive-macros at all because I realized the initial approach was incorrect. We need to read NULL values and keep them in "tags," but with #[allow_null_container_values], we lose them after parsing the JSON. Please review the test I've added; I tried to be very detailed and include some kind of contract there.

@scovich So few more findings after your comment.

but we will also need to update the row visitors because they don't support nullable map keys today

That's right, but we don't read tags at all: https://github.com/delta-io/delta-kernel-rs/blob/main/kernel/src/actions/visitors.rs#L120.
So, changing the schema of AddAction shouldn't break anything in RowVisitor?

Second thing. I also realized that other actions also have "tags":

• Remove: https://github.com/delta-io/delta-kernel-rs/blob/main/kernel/src/actions/mod.rs#L786. This one is the same as Add and also nullable in Delta-java: https://github.com/delta-io/delta/blob/master/kernel/kernel-api/src/main/java/io/delta/kernel/internal/actions/RemoveFile.java#L40
• CheckpointMetadata, Cdc, Sidecar

@scovich Hey Ryan, can you please take a look at my response above, please: #1395 (comment)?

If I'm missing something, or if changing the tags' schema doesn't make sense without supporting them in RowVisitor, I can also try adding parsing tags into RowVisitor. I just don't want to merge everything into a single PR if it can be two separate parts, but if you think it's better to combine both parts into one PR, I will do that!