netlink: test debug output on big-endian systems

[dlanov]

Part of #278.

TestNlmsgFprintf used captured nftables netlink messages whose
attribute headers were encoded for a little-endian system. Netlink
attribute headers use native byte order, so those fixtures produced
invalid lengths and an out-of-bounds panic when the test ran on s390x.

Build the test messages dynamically with AttributeEncoder instead:

• netlink attribute headers are encoded in native byte order
• nftables integer payloads are encoded in network byte order
• named NFTA_* and NFT_* constants describe the test rule
• semantic helpers construct the nested nftables expressions

This removes the runtime endian conversion and allows the existing
big-endian skip to be removed without changing production behavior.

Add s390x coverage using QEMU user-mode emulation. Rename
TestReceiveIter to TestIntegrationReceiveIter so live-kernel tests
can be excluded consistently with -skip Integration, while the root,
nlenc, and nltest packages run under QEMU.

Testing:

• go test -run '^TestNlmsgFprintf$' -count=1 -v
• GOOS=linux GOARCH=s390x CGO_ENABLED=0 go test -exec qemu-s390x-static -tags gofuzz -count=1 -skip Integration ./nlenc ./nltest .
• go test ./... -count=1
• go vet ./...
• git diff --check

[nickgarlis]

Hi @dlanov

Thanks for the contribution! I've left a couple of comments on my thoughts on the PR.

cc @aojea who implemented the mnl debugger in 5c0c409

[aojea]

clearly I show this area of different architecture is not my forte :)