Harden `launch` server: bind 127.0.0.1 only + validate scan targets

[jaurakunal]

Context

Two CodeQL alerts were dismissed as by-design because isitsecure is a local, user-directed CLI — the operator supplies both the scan target URL and the repo path, so there is no trust boundary crossed:

• py/full-ssrf in engine/shared/rate_limited_client.py — the DAST HTTP client requests the user's own target URL; that's its core function.
• py/path-injection in engine/code_analysis/repo_ingestion.py — the path comes from the user's own --repo argument.

The one scenario where they would matter

Both become real if the launch server is ever bound beyond localhost or run multi-user, letting a remote party direct scans at arbitrary internal hosts/paths.

Defense-in-depth to add

• Bind the launch server to 127.0.0.1 by default; require an explicit opt-in flag to bind any other interface.
• When a scan is triggered over HTTP (not the CLI), validate targets: reject link-local / metadata (169.254.0.0/16), loopback-to-internal pivots, and non-file:// local paths outside an allowed root.
• Document the local-only threat model in the README/SECURITY.md.

Not a blocker for single-user local use (the supported mode today); tracked so the dismissals stay honest.