feat(server): add asgi_middleware param to serve()

[bokelley]

Closes #415

Adds asgi_middleware to serve(), _serve_mcp(), _run_mcp_http(), and _serve_a2a(). The parameter accepts a sequence of entries that are each either a (callable, kwargs) tuple or a bare callable factory f(app) -> app. Both forms compose outermost-first. Applied on HTTP transports (streamable-http, sse, a2a); silently ignored (with a logger.warning) on stdio.

A module-level ASGIMiddlewareEntry type alias is added and exported from adcp.server, mirroring the SkillMiddleware pattern.

What was tested

• pytest tests/test_serve_asgi_middleware.py — 6 new tests: no-op, outermost-first ordering, kwargs passthrough, callable factory, functools.partial factory, mixed tuple+callable ordering. All pass.
• Full suite (pytest tests/ --ignore=tests/integration) — 2318 passed, 21 skipped, 1 xfailed (pre-existing network test excluded).
• ruff check src/adcp/server/serve.py src/adcp/server/__init__.py — clean.
• mypy src/adcp/server/serve.py — no new errors (pre-existing missing-stub errors for optional deps are unaffected).

Pre-PR review

• code-reviewer: approved — no blockers; noted stdio silent-drop (fixed with logger.warning) and commit title accuracy (addressed).
• dx-expert: approved after one blocker fix — widened type → Callable[..., Any] in the tuple annotation so functools.partial as the tuple's first element passes mypy. Also flagged missing ASGIMiddlewareEntry export (fixed).

Remaining nits (surfaced, not fixed):

• _apply_asgi_middleware has no validation guard for malformed entries (e.g. 3-tuples produce a decontextualized ValueError). Follow-up issue welcome.
• No test for the size_limit ordering invariant (_wrap_with_size_limit always outermost). Worth a follow-up.
• create_mcp_server() auth docs reference add_middleware() as an alternative pattern without cross-referencing asgi_middleware=. Docs follow-up.

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_01GttQNCHuVSVUyRW5knbitw

---

Generated by Claude Code