Normalize Read result for empty collections

[t0yv0]

This workaround is aimed to reduce dirty refresh incidence on diffs between empty and null collections that is benign in TF but projects worse to Pulumi.

[codecov]

Codecov Report

Attention: Patch coverage is 58.33333% with 10 lines in your changes missing coverage. Please review.

Project coverage is 61.20%. Comparing base (b012fa8) to head (437997b).

Files	Patch %	Lines
pkg/tfshim/sdk-v2/provider2.go	58.33%	9 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2069      +/-   ##
==========================================
- Coverage   61.21%   61.20%   -0.01%     
==========================================
  Files         337      337              
  Lines       45177    45200      +23     
==========================================
+ Hits        27654    27666      +12     
- Misses      16003    16013      +10     
- Partials     1520     1521       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[t0yv0]

At this point it's making TestAccBucketPy pass dirty refresh check in AWS. The change has to be scoped down to PlanResourceChange path because rolling it out generally in the bridge introduces more dirty refresh than we had before. This is pretty interesting, we might need to go super-surgical.

@VenelinMartinov suggested the correction may only be needed for maps, let me check lists and sets shortly.

[t0yv0]

What's the best way to decide if it's a block or attr these days @VenelinMartinov

[VenelinMartinov]

You can get the attributes of a resource from res.CoreConfigSchema().Attributes

[VenelinMartinov]

Something like this:

	sch := res.CoreConfigSchema()
	for key, attr := range sch.Attributes {
		if !attr.Type.IsCollectionType() {
			continue
		}

		if config.IsSet(key) {
			continue
		}

		stateVal, ok := m[key]
		if !ok {
			continue
		}
		if !(stateVal.Type().IsCollectionType() && !stateVal.IsNull() && stateVal.LengthInt() == 0) {
			continue
		}

		m[key] = cty.NullVal(attr.Type)
	}

[VenelinMartinov]

I've continued this in #2065

[t0yv0]

Guessing based on the investigation in #2063 that this needs to be scoped to non-computed collections. The best end to end test here would be to cover the idea that we get a clean refresh for these attributes.

[VenelinMartinov]

We likely need to handle this for every block, right

[t0yv0]

Sounds like it, but I don't have an e2e example handy.

[t0yv0]

Closing in favor of #2073