Make drain delay configurable via drainDelaySeconds field

[mt-bt]

Summary

• Adds an optional drainDelaySeconds field to RestateSpec so the delay before removing old drained versions can be configured
• Defaults to 300 seconds (5 minutes) for backward compatibility
• Applies to both ReplicaSet and Knative deployment modes

Motivation

There was already a // todo configurable? comment on the hardcoded TimeDelta::minutes(5) in replicaset.rs.

In resource-constrained environments, a shorter drain delay lets the operator evict old pods more quickly, freeing capacity for incoming ones.

[tillrohrmann]

Thanks for creating this PR @mt-bt. We are going to take a look at it. cc @lukebond for visibility.

[lukebond]

thanks for this @mt-bt, really clean change. it's a purely additive optional field with nullable: true and skip_serializing_if, so existing RestateDeployment resources are completely unaffected since they'll just get the 300s default; appreciated.

a few things before merging:

1. crd/RestateDeployment.pkl file needs updating to match. add drainDelaySeconds: Int? to the Restate class around line 67, alongside useHttp11. you can look at how useHttp11 was added there as a reference. then run just generate-examples to make sure everything stays in sync.
2. we have a release notes process now, please add a file at release-notes/unreleased/96-configurable-drain-delay.md. see release-notes/README.md for the format, the "New Feature Example" section is probably closest to what you need here.
3. in knative.rs you nicely log the drain delay when scheduling removal, but the equivalent code path in replicaset.rs doesn't log anything. could you add a similar info!() line there so both deployment modes have consistent logging?

for the CLA you just need to comment on the PR with the exact text:

I have read the CLA Document and I hereby sign the CLA

(CLA Document is here)

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[mt-bt]

I have read the CLA Document and I hereby sign the CLA

[mt-bt]

@lukebond thanks for the review! I've addressed all three points:

• Updated crd/RestateDeployment.pk with drainDelaySeconds
• Added release notes at release-notes/unreleased/96-configurable-drain-delay.md
• Added info! logging with structured fields in replicaset.rs to match knative.rs

When running just generate-example, I get the following diff, I suspect it is not related to this PR and might have been out of sync already? I didn't check this diff in, but let me know if you do want it in.

+++ b/crd/examples/restatecluster.yaml
@@ -3,11 +3,7 @@ kind: RestateCluster
metadata:
  name: restate-test
spec:
-  cluster:
-    autoProvision: true
  compute:
-    image: restatedev/restate:1.5.6
+    image: restatedev/restate:1.4
  storage:
    storageRequestBytes: 2147483648
-  config: |
-    auto-provision = false

[lukebond]

@mt-bt thanks. looks good.

that pkl diff suggests the changes in #77 were missing updates to the schema. i'll sort that out separately, we can merge this change as is without adding that generated output as it would remove things.

[lukebond]

@mt-bt i have a couple of other changes lined up and, once merged, i'll do another release