Fix the logic for null[*]

[remilapeyre]

According to https://discuss.hashicorp.com/t/pattern-to-handle-optional-dynamic-blocks/2384/2
null[*] is defined and should return an empty tuple. The logic in
SplatExpr.Value() is slightly wrong thought as it returns early with a
dynamic value even when the value is actually known to be null.

The fix is easy, just check for null sources first, then for dynamic
values. I added a test for this in TestExpressionParseAndValue().

[mildwonkey]

👍 thank you!

[apparentlymart]

Thanks for working on this @remilapeyre!

Based on the discussion over in hashicorp/terraform#26746, the behavior change you've proposed seems correct and consistent with our usual sense of backward compatibility for HCL behaviors of this sort (the new result is no more unknown than the old). I'm going to merge this now.

Because null[*] is not a common language construct and since this has unfortunately narrowly missed an earlier HCL patch release today I'm not going to issue another one just for this one change. There are some other open pull requests currently awaiting review and so I expect at least one of those will prompt a new release before too long. With that said, this does mean that it won't be possible to incorporate this into Terraform immediately because Terraform typically tracks only tagged releases of libraries unless there's a high-urgency reason to do otherwise. We can bring it in to Terraform once it's available in a HCL release, though.

Thanks again!