Illustrate why TestResourcesDataApp fails

[michaeldiamant]

Illustrates why TestResourcesDataApp fails in https://app.circleci.com/pipelines/github/algorand/go-algorand/10048/workflows/dab1223e-f91d-4865-8824-ab00327a41f1/jobs/176490 by reverting https://github.com/algorand/go-algorand/pull/4669/files#diff-f98dee0358238dafbdeae29c761862fbc496f0af2335d85cdec69954a82ba848R214. The PR intends to generate discussion + agree on desired behavior.

Notes:

• With this PR, go test ./ledger -run "TestResourcesDataApp" -count=1000 passes. Without the PR, it consistently fails.
• Prior to this PR, TestResourcesDataApp fails because it assumes a randomly generated AppLocalState contains either a non-zero NumUint or non-zero NumByteSlice.
• Though it's not immediately clear to me if the generator of test design needs to be changed (bolded is my assumption).
  • If Tests: Consistent resource/account struct tests #4669 is correct, then TestResourcesDataApp test design must be modified.
  • Otherwise, the generator should be reverted by merging this PR.

[michaeldiamant]

Closing - Based on further source code inspection, I concluded the generator is correct without this PR and TestResourcesDataApp must be modified.

If folks feel I reached the wrong conclusion, let me know with a note here.

Rationale:

• I see initialization AppLocalState initialization in ledger/apply passes the App's LocalStateSchema:
  

  go-algorand/ledger/apply/application.go

  Lines 245 to 247 in a3e90ad

  	localState := basics.AppLocalState{
  	Schema: params.LocalStateSchema,
  	}

• Since App's must define local and app schemas at initialization, I think it's safe to conclude AppLocalState.Schema is always non-nil. 0 byteslice and 0 int local state usage is a valid use case. Consequently, this PR's change is incorrect because it artificially limits the range of valid examples.

[michaeldiamant]

For completeness - #4699 contains a proposed fix.