Replace external ruleset with curated subset of gitleaks' MIT-licensed rules

[fahadsiddiqui]

Problem

data/regex_list_1.json is a generic third-party DLP sample list (Arista/Cisco configs, US bank routing numbers, Box links, CVE IDs, etc.) with mixed quality — much of it irrelevant to a coding-assistant use case. After the coding-only keyword filter, only 30 of 66 entries survive, and (per #3) some of those are broken. The dataset is ballast rather than a curated asset.

Fix

Replace or supplement data/regex_list_1.json with a converted subset of gitleaks' MIT-licensed ruleset (https://github.com/gitleaks/gitleaks), which is actively maintained, vastly higher precision, and purpose-built for this exact use case (secrets in code/text).

Suggested approach:

• Write a one-time conversion script (scripts/convert-gitleaks-rules.ts or similar) that maps gitleaks TOML rules into this project's external-rule JSON schema
• Preserve the existing stable-ID scheme so config references (disabledRules, etc.) don't break for existing users
• Vet every converted rule through the existing safe-regex2 + empty-match checks before inclusion
• Drop routing-number/networking-config style entries that don't apply to this product's threat model

Acceptance criteria

• data/regex_list_1.json (or its replacement) is derived from gitleaks' ruleset with clear attribution/license notice
• All included rules pass safe-regex2 vetting with zero rejections at load time
• Conversion script is committed so the ruleset can be refreshed later
• Existing tests (rule-loading, FP corpus) still pass