fix(compat): extract hostname from brand_manifest URLs with paths

[bokelley]

Closes #677

Summary

strip_url_scheme only strips the scheme prefix and trailing slash — for a v2.5 brand_manifest URL like "https://acme.com/.well-known/brand.json" it returned "acme.com/.well-known/brand.json", which fails BrandReference.domain's hostname-only regex and surfaced as a confusing INVALID_REQUEST[brand.domain] error. The v2.5 spec documents brand_manifest as a URL to a JSON file, so a path is the expected input shape.

This PR adds extract_brand_domain() to _url.py using urlparse().hostname, which correctly strips both path and port. A bare-domain fallback via strip_url_scheme handles callers that pass a schemeless string. Both adapter call sites — get_products.adapt_request and _media_buy_helpers.adapt_brand_manifest_to_brand (used by create_media_buy) — are updated. update_media_buy is not affected (it carries no brand_manifest).

What was tested

• pytest tests/test_legacy_adapter_v2_5_get_products.py tests/test_legacy_adapter_v2_5_media_buy.py tests/test_dispatcher_shape_detection.py -v — 54 passed
• ruff check src/adcp/compat/ — no issues
• mypy src/adcp/compat/ — no issues in 12 source files

Nits surfaced by pre-PR review (not fixed in this PR)

• Schemeless URL with path (e.g. "acme.com/.well-known/brand.json") falls through to strip_url_scheme and produces an invalid domain — pre-existing gap, not introduced here; v2.5 spec defines brand_manifest as a full URL so this input shape is malformed
• strip_url_scheme retains a public name despite now being an internal fallback inside extract_brand_domain; consider renaming to _strip_url_scheme in a follow-up
• IPv6 literal hostnames are extracted without brackets by urlparse and will still fail BrandReference.domain's regex; documented in extract_brand_domain's docstring, callers must validate if IPv6 is possible (v2.5 spec makes this unrealistic for brand_manifest)

Pre-PR review

• code-reviewer: approved — no blockers; schemeless-with-path fallback gap is pre-existing and documented
• dx-expert: approved — fix is correct at both call sites; nits on strip_url_scheme naming and IPv6 noted above

Triage-managed PR. This bot does not currently iterate on
review comments or PR conversation threads (only on the source
issue). To unblock:

• Push fixup commits directly: gh pr checkout <num> →
  fix → push.
• Or re-trigger: comment /triage execute on the source
  issue.

See adcp#3121
for context.

Session: https://claude.ai/code/session_01YX14H75zYWQ1BYuopyUL34

---

Generated by Claude Code

[bokelley]

Promoting out of draft — CI green across 3.10–3.13, both expert reviews approved, regression tests in place for both get_products and create_media_buy adapter call sites.

Follow-up nits (not blocking this PR):

1. strip_url_scheme naming. Now that extract_brand_domain is the canonical helper for "URL → hostname", strip_url_scheme is reduced to the bare-domain fallback path. The name is misleading for the role it plays today; worth renaming or inlining in a follow-up.
2. IPv6 silent failure. urlparse("https://[::1]/path").hostname returns "::1", which fails BrandReference.domain's regex without a clear error. Acceptable for now (brand_manifest URLs in the wild are hostname-based, not IPv6 literals), but if we ever see this in production we should validate the result and raise a typed error rather than punting to the regex.

Tracking both as ideas; happy to file follow-up issues if useful.

[bokelley]

Thanks for the promotion and the CI confirmation. Both nits are worth tracking — please do file follow-up issues for them. Brief suggested scopes:

1. strip_url_scheme rename — _strip_url_scheme (or inline into extract_brand_domain) to match its reduced, internal role.
2. IPv6 validated error — when urlparse().hostname returns a bare IPv6 literal, validate and raise a typed error rather than letting it fall through to the regex.

Neither is blocking; filing them now gives the next triage sweep a handle to act on them.

---

Triaged by Claude Code.

---

Generated by Claude Code

[bokelley]

Expert review summary

code-reviewer: approved — no blockers. Confirmed urlparse().hostname is the right primitive; tests cover regression at both call sites. Minor nits: a bare-host-with-path (acme.com/path, no scheme) falls into strip_url_scheme and round-trips unchanged — fails downstream regex, which is acceptable fail-fast behaviour for an out-of-spec input.

ad-tech-protocol-expert: approved — confirmed the v2.5→v3 mapping is the only one v3's BrandReference schema permits (additionalProperties: false, domain-only). Surfaced two real spec-level follow-ups (non-blocking, filed as separate issues):

• v2.5 brand_manifest with non-standard path is silently flattened to domain only #683 — v2.5 manifests at non-/.well-known/brand.json paths get silently flattened to {domain}; v3 sellers will fetch the canonical path and 404. Spec-level information loss the adapter cannot avoid; worth a log-warning.
• v2.5 brand_manifest as inline object (not URL) is unhandled by legacy adapter #684 — v2.5 brand_manifest oneOf[object, URL] is real; the adapter only handles the URL branch. Pre-existing gap surfaced during review.

CI: 16/16 green. Merging.