L2s: return a concrete physical name/ARN instead of token when possible

[eladb]

For things like cross-region/account references (#49), it might be needed to use "well-known" physical names for resources when they are referenced across stacks.

Today, L2 properties that represent attributes are always resolved to the L1 attribute. For example, bucket.bucketName will always resolve to { "Fn::GetAtt": [ "ID", "Name" ] }. However, if users explicitly specified a bucket name when they defined the bucket (new Bucket(this, 'Mybucket', { bucketName: 'foo' })), then why not just return "foo" for bucket.bucketName?

Then, consumers can use something like unresolved(bucket.bucketName) to determine if this is a concrete valuer or not when they read the value.

[skinny85]

I believe the current reason for always using { "Fn::GetAtt": [ "MyResource", "Name" ] } instead of just 'foo' is that the former establishes an implicit dependency ordering in CloudFormation. This way, we're certain resources in the Stack will get created in the correct order.

[eladb]

Duplicate with #4093