fix: clean up malformed CNI error output

[immanuwell]

This fixes a small but real logging bug in cni-plugin.

Some CNI error paths use %e for error, so Go prints %!e(...) junk instead of the actual message. That shows up on normal failures like bad CNI JSON, bad kube config, or missing integration mounts. Kinda noisy, and it hides the real problem.

What changed

• switch those log and test format strings to %v or %s
• add a small regression test for the invalid JSON path
• fix the proxy-gid mismatch message in the integration test helper

Repro

1. On main, run go test ./cni-plugin/integration/tests/flannel -run TestLinkerdIsLastCNIPlugin -v
2. In a plain local env, the expected mount is missing, so the test fails
3. Before this patch the failure contains %!e(...)
4. With this patch it prints the real path error

Checks

• go test ./cni-plugin ./proxy-init/... ./pkg/...

[raykroeker]

Thanks @immanuwell for the PR. Instead of using format strings, please use the structured error field on the logger.

[immanuwell]

@raykroeker done, addressed all the requested changes, PTAL

[alpeb]

Thanks @immanuwell , the main change looks good. I don't think you should add a unit test for logging though; it tests the logging framework not our behavior and I can see this sort of thing can become a burden to maintain when the framework changes.