test: improve CLI parsing and source-tracking coverage

[abtreece]

Context

PR #609 adds targeted regression coverage for #601, specifically around config precedence when CLI or environment values are explicitly set to their defaults. While reviewing that work, a few broader CLI test gaps became clear that are adjacent to the fix but not required for it.

Suggested test improvements

• Replace stale default-node tests that manually simulate old Run() behavior with tests that exercise the real applyBackendDefaults() path and backend Run() wiring.
• Add table-driven tests for config source tracking from Kong:
  • global CLI flags
  • global env vars
  • representative backend flags such as --node, --scheme, --file, --filter, and Secrets Manager/ACM flags
  • aliases where TOML and CLI names differ, such as TOML nodes versus CLI node
• Add broader config precedence tests for representative settings beyond the refactor: replace manual config precedence merging #601 regression cases:
  • boolean options with explicit false/true behavior
  • string options
  • duration options
  • backend-specific options
• Keep parse-only CLI tests separate from config-merge tests so failures identify whether Kong parsing, source tracking, or TOML resolution regressed.

Acceptance criteria

• CLI parse/source tests cover representative global and backend flags.
• Config precedence tests cover representative scalar, bool, duration, slice, and backend-specific options.
• Stale tests that only duplicate implementation details are removed or rewritten to exercise production behavior.
• go test ./cmd/confd passes.