fix: streamline discovery targeting and live catalog loading

[dannysteenman]

Summary

• What changed?
  • Reduced AWS live discovery pressure by batching Resource Explorer requests, adding throttling-aware retries, and reusing memoized dataset loads where possible.
  • Redesigned CLDBRN-AWS-CLOUDWATCH-2 to evaluate log groups from latest stream activity summaries instead of enumerating every log stream.
  • Added SDK/provider debug tracing, restored the CLI discover --region single-region flag, and kept multi-region discovery support in the SDK.
• Why was this needed?
  • Broad discovery runs were easy to throttle in Resource Explorer, large CloudWatch log-stream inventories could become very slow, and the CLI targeting/debug surface needed to be simpler and more predictable.

Diagram

flowchart LR
  CLI["CLI discover --region"] --> SDK["CloudBurnClient / SDK"]
  SDK --> RE["Resource Explorer batched catalog"]
  RE --> Datasets["Memoized dataset loaders"]
  Datasets --> CW["CloudWatch latest-stream activity summary"]
  Datasets --> Rules["Discovery rules"]
  CW --> Rules

Loading

Scope

• cloudburn (cli)
• @cloudburn/sdk
• @cloudburn/rules
• docs/community files

Release Notes

• Added a .changeset/*.md file for published package changes
• No published package changes in this PR

Verification

• pnpm lint
• pnpm typecheck
• pnpm test
• pnpm build
• pnpm verify

Boundary Checks

• No engine/parser/provider logic added to @cloudburn/rules
• CLI delegates scan logic to SDK
• README/CONTRIBUTING/docs updated when behavior changed