feat: add host-network mode for Redis templates

[joshrotenberg]

What

Adds host networking mode to the Redis templates, exposing network_mode(mode) and host_network() builder methods on both RedisTemplate and RedisClusterTemplate.

Why

RunCommand already supports network, but templates did not expose host mode. RedisClusterTemplate always created its own bridge network and therefore needed the --cluster-announce-ip ceremony for host-side clients. On Linux, --network host eliminates that ceremony entirely because container ports are real host ports.

How

RedisTemplate (single container):

• host_network() / network_mode("host") set --network host.
• In host mode, published port mappings are skipped (Docker ignores -p when sharing the host namespace); the Redis port is reachable directly on the host (6379 by default).

RedisClusterTemplate (multi-node):

• host_network() / network_mode("host") enable host mode (network_mode only special-cases "host"; any other value keeps the default managed bridge).
• In host mode the template:
  • skips creating and removing the private bridge network,
  • runs every node with --network host and no -p mappings,
  • gives each node a distinct listening port (port_base + index) so nodes do not collide in the shared host namespace,
  • skips the --cluster-announce-ip ceremony, and
  • wires redis-cli --cluster create, node readiness polling, node_role, and cluster_info to reach nodes on 127.0.0.1 at their distinct ports.
• RedisClusterConnection::from_template and node(i).host_port already report port_base + index, which equals the real host port in host mode, so external clients connect to localhost:<port_base+index> with no changes.

Failure behavior on non-Linux

Host networking is a Linux-only Docker feature. On Docker Desktop for macOS/Windows the daemon runs in a Linux VM, so --network host binds inside that VM and is effectively a no-op. The methods do not return an error on non-Linux (the Docker CLI accepts the flag regardless of backend); this is documented loudly in the rustdoc, and the choice (no hard error, document the no-op) is stated explicitly.

Evidence / test plan

• Unit tests (run everywhere, no Docker): basic-template --network host wiring + no --publish; cluster uses_host_network, node_internal_port, node_cluster_address, and build_ping_args for both bridge and host modes, including with a password.
• Integration test test_redis_cluster_host_network_wiring_api asserts host-mode wiring through the public API (distinct host ports, host_port accessor, network_mode("host") == host_network()); runs without Docker.
• Integration test test_redis_cluster_host_network_smoke_test is #[ignore] and gated to Linux via a runtime skip; the CI Docker Integration job (Ubuntu) exercises the Linux host path. It starts a real host-mode cluster, asserts it converges with no bridge network and no announce-ip, and cleans up.

Validation gate run locally (macOS):

• cargo fmt --all
• cargo clippy --all-targets --all-features -- -D warnings
• cargo clippy --all-targets --no-default-features -- -D warnings
• cargo test --lib --all-features (819 passed)
• cargo doc --no-deps --all-features
• cargo test --doc --all-features (416 passed)

Closes #247.