feat: add residual PII audit workflow

[Keerti707]

------Summary-------------------

Adds a lightweight residual PII audit workflow for checking whether redacted JSONL training data still contains sensitive patterns before being used for downstream SFT or DPO training.

This contribution focuses on the privacy-auditing portion of the project requirements and provides a small reusable utility for validating residual-risk after redaction.

------What’s included------------------

• JSONL residual PII audit utility

• Detection support for:

  • email addresses
  • phone numbers
  • API-key-like strings
  • token-like strings

• Risk-level classification

• Example redacted dataset

• Unit tests with pytest

• README usage documentation

• pytest configuration through pyproject.toml

------Why this contribution-----------------

The repository requirements explicitly mention:

• residual PII auditing
• documented audit workflows
• measurable privacy validation

This PR contributes a focused validation utility around those requirements instead of duplicating broader pipeline/export implementations already being explored in parallel PRs.

-------Verification-------------------

pytest -q
python training_setup_logs/pii_audit.py

Both commands execute successfully locally.