Add targeted Lua tests for low-coverage runtime branches

[levleontiev]

Summary

Add targeted Lua tests to raise coverage in the weakest non-generated runtime modules identified by the new luacov reporting path.

Why

Current non-generated Lua coverage is about 82.28%, but several boundary/error-heavy modules remain materially under-covered.

Current lowest-covered handwritten modules:

• src/fairvisor/decision_api.lua — 58.87%
• src/fairvisor/bundle_loader.lua — 76.79%
• src/fairvisor/utils.lua — 77.05%
• src/fairvisor/saas_client.lua — 78.88%
• src/fairvisor/streaming.lua — 84.91%

Scope

Add tests for the highest-value missing branches, prioritised in this order:

1. decision_api.lua
2. bundle_loader.lua
3. saas_client.lua
4. streaming.lua
5. utils.lua

Target gaps

• decision_api.lua
  • flat JSON fallback parser branches
  • retry/throttle edge paths
  • non-ngx hash/jitter fallback branches
  • error/no-op logging/guard branches
• bundle_loader.lua
  • init/deps wiring
  • queue_event failure path
  • descriptor hint extraction branches
  • signing/HMAC fallback path
• saas_client.lua
  • missing resty.http
  • transport/request_uri failures
  • unhappy-path response handling
• streaming.lua
  • empty data: events
  • termination/error event paths
  • reconcile failure / early-return branches
• utils.lua
  • constant-time compare fallback
  • bit/base64 helper branches
  • nil/false return branches

Acceptance Criteria

• Add focused unit/integration tests for the gaps above.
• Increase meaningful non-generated Lua coverage above the current baseline.
• Keep generated code (src/fairvisor/generated/) out of the success metric.
• All existing unit, integration, and E2E suites still pass.