provider-resources: replace AWS-internal patterns with generic Plugin Framework guidance

[AdamTylerLynch]

The provider-resources skill was written against terraform-provider-aws internals
without saying so: r.Meta().ExampleClient(ctx), conns.Client, tftags.TagsSchema(),
verify.SetTagsDiff, tfresource.NotFound, acctest.ProtoV5ProviderFactories. None
of these exist outside that repository, so the skill actively misled anyone building
any other provider. It also conflicted with its sibling skills (legacy
ComposeTestCheckFunc vs provider-test-patterns' ConfigStateChecks; ProtoV5 vs
ProtoV6; legacy website/docs/ layout vs provider-docs' tfplugindocs layout).

Changes

• States the policy up front: Plugin Framework for all net-new resources/data
  sources; SDKv2 only for maintaining existing resources; migrate via
  terraform-plugin-mux (per the AWS provider's contributor guidance). Removed the
  SDKv2 example code accordingly
• All examples now use a generic examplecloud client stored on the resource via a
  standard Configure() + interface assertions — the shape any provider can copy
• Read removes missing resources from state (tflog.Warn + RemoveResource); Delete
  tolerates already-deleted; consistent error grammar ("creating Widget (%s): %s")
• Added the Plugin Framework id guidance: omit a redundant id when the API has a
  natural identifier; ImportStatePassthroughID
• Testing section aligned with provider-test-patterns: ProtoV6 factories, modern
  ConfigStateChecks, required coverage (_basic + import step, _disappears,
  per-attribute), config style rules (indexed %[1]q, self-contained configs,
  randomized names, no hard-coded values)
• Documentation section now points at tfplugindocs/docs/, deferring to the
  provider-docs skill — resolving the three-way docs-layout conflict
• Two new load-on-demand references distilled from the AWS provider contributor docs:
  • references/design-principles.md — resource = smallest building block, one
    resource ↔ one API, singular vs plural data source semantics,
    relationship/async-task/version modeling
  • references/retries-and-waiters.md — three eventual-consistency failure classes,
    status/wait function pairs, post-create not-found retries, typed-error-only retries

Verification: scripts/validate-structure.sh passes; SKILL.md shrinks 599 → 514
lines; zero references to AWS-internal packages remain.

---

Addresses the acceptance-test coverage, docs-generation, singular+plural data source, changelog, and Framework-first guidance requested in #14 (the follow-up commit adds the terraform-plugin-mux detection heuristic suggested there, the recommendation to ship both data source forms, and a changelog checklist item).

[AdamTylerLynch]

Duplicate/overlap flag: #41 (@bbasata) also rewrites this same file, terraform/provider-development/skills/provider-resources/SKILL.md, and the two PRs take different approaches to the same problem — they will hard-conflict on merge:

• This PR (provider-resources: replace AWS-internal patterns with generic Plugin Framework guidance #79): keeps provider-resources as a single Framework-first skill and strips the terraform-provider-aws-internal patterns (conns.Client, tftags, verify.SetTagsDiff, ProtoV5, legacy website/docs/).
• Refactor provider-resources skill into SDK-specific guidance skills #41: splits provider-resources into a router + two new skills (provider-resources-with-framework, provider-resources-with-sdk).

Both are answers to the "should this be split into Framework/SDKv2 skills?" discussion in #14 (@ffalor / @bbasata). Flagging so maintainers can reconcile the two rather than merge one and silently clobber the other. My related #86 (provider-framework-migration) is where I moved the SDKv2 content, which keeps this skill Framework-only without needing the SDK-specific split.

[leefowlercu]

Thanks for the heads up @AdamTylerLynch. Will update this thread when resolution is decided.