Mob.DNS: Android in-process IPv4 resolution via Bionic getaddrinfo NIF

[GenericJam]

What

Adds an Android nif_resolve_ipv4 so Mob.DNS.resolve/1 and preresolve/1 work
on Android the same way they do on iOS (PR #32). The Elixir surface is unchanged
— no platform branching in caller code.

Why

:inet.getaddr/2 returns :nxdomain on physical Android devices we deployed
to, even though the same app's in-process HTTPS stack resolves those hostnames
fine. Verified on a Moto G Power 5G 2024 (Android 14); the Android emulator
doesn't hit this (its DNS proxy at 10.0.2.3 is reachable to any resolver),
which is why it didn't surface earlier.

BEAM's default DNS path forks inet_gethost (a port program) and reads what
its getaddrinfo returns. Suspected cause: libnetd_client.so's routing into
netd doesn't survive the execve into a port program — the in-process call
goes through netd binders that pick up the per-network DNS, the execve'd child
sees an empty resolver. I haven't pinned the exact mechanism; happy to take a
follow-up with the actual diagnosis. Either way the NIF sidesteps it by
running getaddrinfo in the app's own process.

What changed

• android/jni/mob_zig.zig — Bionic getaddrinfo / freeaddrinfo /
  addrinfo / sockaddr_in / EAI_NONAME / EAI_NODATA / EAI_AGAIN /
  AF_INET / SOCK_STREAM bindings. Layout mirrors AOSP
  bionic/libc/include/netdb.h exactly — BSD-derived ai_canonname before
  ai_addr in struct addrinfo.
• android/jni/mob_nif.zig — nif_resolve_ipv4 mirroring iOS's in
  ios/mob_nif.m (same return atoms :nxdomain / :timeout / :no_address /
  {:gai, code}, same ERL_NIF_DIRTY_JOB_IO_BOUND scheduling). Registered in
  nif_funcs[].
• lib/mob/dns.ex — moduledoc updated: dropped the "Android isn't
  affected" claim; added a background-app caveat (Android App Standby blocks
  all outbound network from a backgrounded mob app — TCP-by-IP, not just
  DNS — surfaces as :closed/:timeout on any socket; fix is a foreground
  service or keep the app foregrounded).
• common_fixes.md — new section documenting the symptom + fix so the
  next person who sees :nxdomain on a physical Android finds the workaround.

Verified

End-to-end on the Moto G via RPC (BEAM PID matches deployed APK; `Mob.DNS` is
the only DNS path used):

```elixir
Mob.DNS.resolve("repo.hex.pm") #=> {:ok, {151, 101, 21, 91}}
Mob.DNS.resolve("google.com") #=> {:ok, {173, 194, 43, 113}}
Mob.DNS.resolve("nonexistent.invalid") #=> {:error, :nxdomain}
:inet.getaddr(c"repo.hex.pm", :inet) #=> {:ok, {151, 101, 21, 91}} (via seeded :file)
Mix.install([{:short_uuid, "> 0.1"}]) #=> resolves, fetches, compiles on-device
ShortUUID.encode(uuid) #=> "MCoJqPMVQiCPfeahTVNigE"
```

Quality

• mix test: 804 + 27 doctests pass, 0 failures.
• mix credo --strict on lib/mob/dns.ex: clean.
• mix format --check-formatted: clean.
• zig fmt --check on mob_nif.zig + mob_zig.zig: clean.

Notes for the reviewer

• Same atom vocabulary + dirty-IO scheduling as the iOS NIF in PR DNS: document preresolve as the robust iOS path (cellular-safe) #32, so
  callers see identical error shapes across platforms.
• I didn't add an Android-specific Elixir test — test/mob/dns_test.exs
  already covers the public surface and handles the :nif_not_loaded
  off-device case. On-device NIF behavior isn't exercised in the host test
  suite (same as iOS). If we want device-tier tests later that's a separate
  follow-up.
• Compile warnings about before_closing_body_tag/1 in mix.exs are
  pre-existing and unrelated to this patch.

🤖 Generated with Claude Code