I would be curious to compare the schema generated by this method vs the one generated by static TFGen for something large like AWS, not to check it in but to grasp the discrepancies. I suspected historically that there is some information we're not able to recover from the gRPC protocol possibly including
whether a property is optional or not
whether a proeprty is MaxItems=1 or not
This would be great to understand and document, as obviously a gRPC schema loading approach is preferable, if the limitations aren't there or can be worked around we could consider further leaning into it.
Follow up on https://github.com/pulumi/pulumi-terraform-bridge/pull/1973#discussion_r1629758340: