Add explicit prefix-based secret token detection

[Jeffrin-dev]

Motivation

• The existing detector relied on entropy heuristics and optional detect-secrets plugins which missed prefix-style tokens such as modern OpenAI keys (e.g. sk-proj-...).
• Deterministic prefix-based regexes are more reliable for common token formats and should be evaluated before entropy scoring and independent of detect-secrets availability.

Description

• Added PREFIX_PATTERNS to SecretsDetector with regexes for sk- (OpenAI), ghp_ (GitHub), AKIA (AWS access keys), and xoxb- (Slack bot tokens).
• Implemented _detect_prefix_patterns and wired it so prefix matches are collected and merged before library-based and entropy-based detections.
• Updated and added parametrized tests in tests/test_policy_and_secrets.py to cover all four explicit patterns.

Testing

• Ran pytest -q tests/test_policy_and_secrets.py and all tests passed (8 passed).

---

Codex Task