feat: security review gate for agent-generated PRs

[BnJam]

Summary

The technocore review pipeline should automatically detect and close PRs containing
malicious code before a human ever sees them. This is particularly important for
agent-generated PRs where the implementing agent could theoretically be prompt-injected
via malicious issue content.

Patterns to detect

• Credential exfiltration: tokens/keys written to files, env vars sent to external hosts
• Backdoors: reverse shells, remote eval/exec of downloaded content
• Supply chain attacks: lockfile tampering, non-canonical registries, typosquatted deps
• CI/CD tampering: workflow changes that add unexpected secret access or external runners
• Obfuscated code: heavily encoded strings, steganographic payloads

Proposed behaviour

1. Reviewer agent runs a security gate before any code review
2. On detection: close the PR, add security/malicious label, open a security issue
3. On clean: proceed with standard correctness/maintainability review

Context

This is implemented in the eo-reviewer agent persona in technocore. This issue tracks
the feature so it can be reviewed, refined, and extended as real cases emerge.