ci(codeql): teach analyzer about ValidateSafeURL SSRF barrier

Closes CodeQL alert #23 (go/request-forgery, Critical) at the
structural level — by telling CodeQL what the runtime code already
does — rather than via per-line `// codeql[...]` suppressions.

Background. internal/service/scep_probe.go:232 calls client.Do(req)
where the request URL is built from operator-supplied input. The
runtime defense is two-layer:

  1. validation.ValidateSafeURL(rawURL) at scep_probe.go:86 rejects
     non-http(s) schemes, empty hosts, literal-IP hosts in reserved
     ranges (loopback, link-local incl. cloud metadata
     169.254.169.254, multicast, broadcast, unspecified, IPv6
     link-local), and DNS names whose A/AAAA resolution returns any
     reserved IP. RFC 1918 is intentionally NOT blocked — see
     internal/validation/ssrf.go:17-21 for the design rationale.

  2. validation.SafeHTTPDialContext on the http.Transport (line 254)
     re-resolves at dial time, applies the same reserved-IP set, and
     pins the dial to a literal non-reserved IP — defeating DNS
     rebinding between validate and dial.

CodeQL's go/request-forgery query is a syntactic taint-tracking rule
with no built-in knowledge of either validator, so it reports the
finding even though the runtime is correctly defended.

The fix. Add a Models-as-Data (MaD) extension at .github/codeql/
declaring ValidateSafeURL as a request-forgery barrier. The barrier
applies to Argument[0] (the URL parameter), which means the analyzer
treats every URL flowing through ValidateSafeURL as sanitized for the
request-forgery taint set. After this lands:

  - Alert #23 dismisses at scep_probe.go:232.
  - The same model applies to the second site of this exact shape —
    webhook notifier's outbound client.Do (internal/connector/
    notifier/webhook/webhook.go) — without per-line annotations.
  - Future code that flows operator URLs through ValidateSafeURL
    inherits the barrier automatically.

This is the structural fix, not a band-aid:

  - Band-aid (rejected): `// codeql[go/request-forgery]` suppression
    on line 232. Suppresses one alert; doesn't teach the analyzer.
    Webhook notifier would need the same comment when its sibling
    rule landing fires.

  - Structural (this change): teach CodeQL via models-as-data, in
    config checked into the repo, that lives next to the workflow
    that uses it. The validators ARE sanitizers in the runtime —
    this PR makes the analyzer's model match reality.

Files:

  - .github/codeql/qlpack.yml — local model pack manifest, declares
    extensionTargets: codeql/go-all: '*'

  - .github/codeql/models/request-forgery-sanitizers.model.yml —
    barrierModel row for validation.ValidateSafeURL Argument[0] /
    request-forgery taint kind / manual provenance

  - .github/codeql/codeql-config.yml — references the local pack +
    keeps security-and-quality query suite scope

  - .github/workflows/codeql.yml — Initialize CodeQL step picks up
    config-file: ./.github/codeql/codeql-config.yml. The existing
    `queries: security-and-quality` line stays so even if the config
    file fails to load, the suite scope is preserved.

  - docs/architecture.md::Input Validation and SSRF Protection —
    extended to name the egress validators (ValidateSafeURL +
    SafeHTTPDialContext) and the call sites (SCEP probe + webhook
    notifier). Closes the docs gap surfaced during the audit; the
    egress threat-model previously lived only in source comments.

Requires CodeQL CLI ≥ 2.25.2 for the barrierModel extensible
predicate (Go MaD support added 2026-04-21). github/codeql-action@v3
ships a recent enough CLI by default; if a future analysis fails
with "unknown extensible predicate barrierModel", the action's CLI
has regressed below 2.25.2 — pin a newer action version rather than
reverting this pack. Documented inline in qlpack.yml.

References:
  - https://codeql.github.com/docs/codeql-language-guides/customizing-library-models-for-go/
  - https://github.blog/changelog/2026-04-21-codeql-now-supports-sanitizers-and-validators-in-models-as-data/

docs/architecture.md

@@ -993,6 +993,8 @@ Audit recording is async (via goroutine) so it never blocks the HTTP response. I
 
 All shell-facing inputs (connector scripts, domain names, ACME tokens) are validated through `internal/validation/command.go` before reaching shell execution. `ValidateShellCommand()` denies all shell metacharacters. `ValidateDomainName()` enforces RFC 1123. `ValidateACMEToken()` restricts to base64url characters. The network scanner filters reserved IP ranges (loopback, link-local including cloud metadata 169.254.169.254, multicast, broadcast) to prevent SSRF, while preserving RFC 1918 private ranges for legitimate internal scanning.
 
+**Outbound HTTP egress** is defended in two layers, both implemented in `internal/validation/ssrf.go`. Layer one is `ValidateSafeURL(rawURL)`, called at the surface of every outbound HTTP path that accepts an operator-supplied URL. It rejects non-`http(s)` schemes, empty hosts, literal-IP hosts in reserved ranges (loopback `127/8` + `::1`, IPv4 link-local `169.254/16` including AWS/Azure/GCP cloud metadata at `169.254.169.254`, IPv4 multicast `224/4`, IPv4 broadcast `255.255.255.255`, unspecified `0.0.0.0` + `::`, and IPv6 link-local `fe80::/10` + IPv6 multicast `ff00::/8`), and DNS names whose `A`/`AAAA` resolution returns any IP in that set. Layer two is `SafeHTTPDialContext(timeout)`, installed on the `http.Transport.DialContext` of every long-lived outbound HTTP client — it re-resolves the target host at dial time, applies the same reserved-IP set, and pins the TCP dial to a literal non-reserved IP so a racing DNS rebind cannot redirect the connection between validation and dial. RFC 1918 ranges (`10/8`, `172.16/12`, `192.168/16`) are intentionally NOT treated as reserved — certctl is designed to manage certificates inside private networks, and filtering them would break the primary use case. Current call sites: the SCEP probe in the network scanner (`internal/service/scep_probe.go`) and the webhook notifier (`internal/connector/notifier/webhook/`); both wire `ValidateSafeURL` at config + request entry and `SafeHTTPDialContext` on their dedicated `http.Client`. The egress validators are also declared as request-forgery barriers in the CodeQL Models-as-Data extension at `.github/codeql/models/`, so the static analyzer's view of the dataflow matches the runtime defense.
+
 ### Request Body Size Limits
 
 All incoming HTTP request bodies are capped by `http.MaxBytesReader` middleware (default 1MB, configurable via `CERTCTL_MAX_BODY_SIZE`). Requests exceeding the limit receive a 413 Request Entity Too Large response. The middleware is positioned before authentication in the chain so oversized payloads are rejected early, before any auth processing or database work occurs. Requests without bodies (GET, HEAD, nil body) skip the limit check.