feat(supervisor): program bypass firewall rules via nf_tables netlink instead of nft

[akram]

Problem Statement

The proxy-mode bypass-detection firewall rules are installed by shelling out to the nft binary — 6 call sites (run_nft) in crates/openshell-supervisor-process/src/netns/mod.rs — resolved from the workload image (NFT_SEARCH_PATHS). A bare Alpine image ships no nftables, so rule installation cannot run without the community base image's tools.

Proposed Design

Program the nf_tables ruleset directly over netlink from the supervisor (e.g. rustables or nftnl-rs), producing the same table/chain/rule set the current nft invocations create, with no nft process spawned. Rule installation stays best-effort/non-fatal exactly as today (failures reported via OCSF).

Acceptance Criteria

• Bypass-detection rules are programmed without invoking nft.
• The resulting ruleset and enforcement behavior are equivalent to the current nft path.
• Verified on Docker, rootless Podman, and Kubernetes.
• No behavior change for images that already ship nftables.

Scope

In: the nft bypass-rule installation in netns/mod.rs.
Out: the ip/nsenter namespace setup (sibling issue) and dmesg→NFLOG bypass monitoring (#2382).

Related

• Sub-task of feat(supervisor): netlink/syscall network setup so the privileged supervisor needs no workload-image tools #3280 (itself a sub-task of feat(sandbox): make privileged supervisor independent of workload-image code #2750)
• Builds on refactor: replace iptables with nftables for sandbox and VM networking #1335 (nftables migration, which retained privileged CLI execution)

[johntmyers]

Do we need this the changes we have pending? @drew