feat: add hub command to browse and deploy from the runpod hub

[TimPietruskyRunPod]

Summary

• Adds runpodctl hub command group (list, search, get) to browse the RunPod Hub marketplace via the existing GraphQL listings API
• Adds --hub-id flag to serverless create as an alternative to --template-id, enabling one-command deployment from hub repos
• When --hub-id is used, the CLI resolves the listing, extracts the build image and config from the latest release, creates an inline template via the GraphQL saveEndpoint mutation (matching the web UI flow), and deploys the endpoint

New commands

runpodctl hub list                              # top 10 by stars
runpodctl hub list --type SERVERLESS            # only serverless repos
runpodctl hub search vllm                       # search by name
runpodctl hub get runpod-workers/worker-vllm    # get by owner/name
runpodctl hub get <listing-id>                  # get by id

Serverless create from hub

runpodctl hub search vllm
runpodctl serverless create --hub-id cm8h09d9n000008jvh2rqdsmb --name "my-vllm"

GPU IDs and container disk size are automatically pulled from the hub release config. Users can override with --gpu-id.

Test plan

• Unit tests pass (go test ./...)
• E2E tests for hub list, search, get (9 tests)
• E2E tests for serverless create validation (2 tests)
• E2E test for serverless create from hub with cleanup (1 test)
• All 12 new e2e tests pass against live API
• Docs regenerated

[promptless]

📝 Documentation updates detected!

Updated existing suggestion: Update runpodctl documentation for v2.0

---

Tip: Create additional Docs Collections to keep multiple repositories in sync 📚

[max4c]

Looks good — clean command structure, solid E2E coverage, and good UX touches (owner/name format, helpful error messages). A couple minor suggestions for follow-up: (1) document that --type filtering is client-side so limit may return fewer results than expected, (2) check for double error output from output.Error + RunE return, (3) consider resolving the hardcoded listing ID in TestCLI_ServerlessCreateFromHub dynamically. None are blockers.

[TimPietruskyRunPod]

thanks for the review @max4c! addressed all three points in 351603f:

1. --type filtering docs — added note to both hub list and hub search flag descriptions: "applied client-side; --limit may return fewer results"

2. double error output — this is actually fine by design. SilenceErrors = true in cmd/root.go:77 prevents cobra from re-printing the error. the output.Error(err) + return err pattern is intentional across the whole codebase — output.Error emits the structured json error, and the returned error is used for exit code only.

3. hardcoded listing id — TestCLI_ServerlessCreateFromHub now resolves the vllm listing id dynamically via hub get runpod-workers/worker-vllm instead of the hardcoded cm8h09d9n000008jvh2rqdsmb.