revert: kv.Store back to non-durable until CRD properly migrated

In-cluster validation revealed the cluster's TinySignal CRD schema is
missing the EdgeID and From fields the SDK v0.4.0+ writes. K8s API server
silently strips unknown fields, leaving signals unable to be delivered
correctly (no source port → no edge config evaluation → primary key
errors → infinite transient retry → 1400+ signals piled up in 30 min).

Reverting Port.Durable=true here so cluster operation returns to fast
path immediately. Re-enabling durability is gated on:
  1. make manifests regenerating CRD with EdgeID/From fields
  2. Apply the updated CRD to the cluster
  3. Bulk-delete the orphaned poisoned signals
  4. Re-mark Durable=true and redeploy

The SDK durability primitives themselves (v0.4.0/v0.4.1) are correct;
this is an operations gap.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: tpoxa

components/kv/kv.go

@@ -366,13 +366,12 @@ func (c *Component) Ports() []module.Port {
 		{
 			Name:  StorePort,
 			Label: "Store",
-			// Durable: writes are persisted as TinySignal CRDs before
-			// dispatch. Caller's edge unblocks at "work owed", not "work
-			// done". A pod crash mid-handle leaves the signal in the
-			// cluster; the new leader picks it up and retries. Kv writes
-			// are idempotent on (primaryKey, payload), so at-least-once
-			// delivery is safe.
-			Durable: true,
+			// NOTE: Durable=true was attempted in v0.4.0/v0.4.1 but the
+			// cluster CRD schema for TinySignal is missing the EdgeID
+			// and From fields the SDK now writes (CRD wasn't regenerated
+			// + applied). API server silently strips them, reconciler
+			// can't reconstruct edge config, signals pile up unfixable.
+			// Reverted to non-durable until CRD is properly migrated.
 			Configuration: StoreRequest{
 				Operation: OpStore,
 				Document:  settings.Document,