fix: HTTP2 nil pointer panic and broken GOMEMLIMIT under hostPID

[mayankpande88]

Summary

Two fixes found during production health check across all 3 clusters (KP, KR, DEV):

• HTTP2 nil pointer panic: When maxActiveRequests (100) limit is hit in decodeHeaderBlock(), break exits the switch but decoder.Write() still runs with a stale emit func from a previous call, dereferencing a nil req. Fix: set a no-op emit func before breaking so HPACK dynamic table stays in sync without the panic. This was causing 100K+ recovered panics per 6h across KP and DEV, wasting CPU and inflating page cache by ~150Mi per affected pod from log spam.

• GOMEMLIMIT never set under hostPID: readCgroupMemoryLimit() reads /sys/fs/cgroup/memory.max which doesn't exist under the agent's init.scope cgroup (hostPID places PID 1 at the host root). The pod's actual 1Gi limit lives at /sys/fs/cgroup/kubepods.slice/.../cri-containerd-xxx.scope/memory.max. Fix: read /proc/self/cgroup to resolve the container's own cgroup path. GOMEMLIMIT was dead code on all 3 clusters — verified via kubectl exec that memory.max at init.scope doesn't exist. Without GOMEMLIMIT, GC doesn't trigger during ELF parsing spikes (50-100MB per Go binary), letting RSS hit 1Gi → OOM kills.

Evidence

Cluster	OOMKills	Panic count (6h)	Peak memory
KP (prod)	0	35K (lhbc8), 23K (tgtfk), 13K (vjnz9)	-
KR (rackspace)	2	0 (old image)	-
DEV (GKE)	2	39K (v6d55), 29K (fh94b)	1024Mi (v6d55, xdmxm, q9pbt)

Cgroup verification on all clusters:

/proc/1/cgroup  → 0::/init.scope          (host root, no memory.max)
/proc/self/cgroup → 0::/kubepods.slice/... (memory.max = 1073741824)

Test plan

• Verify gofmt -l . passes (checked locally)
• Deploy to DEV, confirm GOMEMLIMIT log line appears at startup
• Monitor DEV panic counts — should drop to zero
• Monitor DEV memory peaks — should stay well below 1Gi
• After 24h stability on DEV, roll to KP

[gemini-code-assist]

Code Review

This pull request improves HTTP/2 header decoding by ensuring the HPACK dynamic table remains synchronized when the maximum number of active requests is reached. Additionally, it updates the memory limit detection logic to correctly identify container limits in hostPID environments by parsing /proc/self/cgroup. Feedback was provided to extend the memory limit detection to support cgroup v1 in addition to v2 for better compatibility across different Kubernetes environments.