fix: normalize confidence case across config filtering and semantic output

[Ty-Robb]

Summary

Normalize confidence casing so threshold filtering behaves consistently across engines, including semantic findings.

Problem

Semantic integration can emit lowercase confidence (for example medium) while threshold checks use uppercase levels (LOW|MEDIUM|HIGH). This can cause findings to be filtered incorrectly.

Changes

• src/config.js
  • Normalize both finding confidence and configured confidence_threshold to uppercase in meetsConfidenceThreshold.
• src/semantic-integration.js
  • Normalize emitted semantic finding confidence to uppercase (MEDIUM default).
• tests/config.test.js
  • Added regression tests for lowercase/mixed-case confidence and thresholds.
  • Added regression test proving lowercase semantic-style confidence is preserved correctly at MEDIUM threshold.

Verification

• npx vitest run tests/config.test.js (pass)
• npx vitest run tests/scan-security.test.js tests/semantic-analysis.test.js (pass)

Fixes #70