Vouch request: crprashant

[crprashant]

What do you want to work on?

I want to fix #4746 , surface k8s scheduling reason for Pending sandboxes. on the Kubernetes driver, if the sandbox pod can't be scheduled
(not enough memory/CPU, taints, no free GPU), sandbox create just sits on
"Requesting sandbox..." and sandbox list/get only say Provisioning. You have
to go to kubectl to find out why, which most hosted-gateway users can't do.

Why this change?

I work at Microsoft and very much interested in contributing to this repo OpenShell.

here are the details of the changes:

• k8s driver: map FailedScheduling events to a "waiting to be scheduled" progress
  state with the scheduler's message, and publish a PodScheduled=False condition
  on the sandbox status while the workload pod is unscheduled. I'm keeping it off
  the Ready condition on purpose, since Ready=False/Unschedulable is treated as a
  terminal Error by the gateway.
• CLI: show "Waiting to be scheduled..." with the reason during create, add a
  "Scheduling:" line to sandbox get, and mark sandbox list as
  "Provisioning (Unschedulable)". JSON output gets it through the existing
  conditions list.
• The wait won't extend the create idle timeout, so a pod that can never fit
  doesn't hang create forever. The timeout error will include the reason.
• No proto or gateway changes needed, and other drivers are untouched.

I have a working branch. I tested it on a k3d cluster: an oversized memory
request shows the exact FailedScheduling message, and when I tainted the node
and then removed the taint, the sandbox waited and then provisioned normally.
Unit tests are added, plus a small e2e test in the Kubernetes lane, and I
updated the sandboxes overview doc.

Checklist

• I wrote this request myself (not AI-generated)
• I have read CONTRIBUTING.md