[History Server] Bump Ray to 2.58.0 in examples and e2e test data

[wuallen57730]

Why are these changes needed?

Task logs don't show up on the task detail page with Ray 2.56, and Ray 2.58 fixes it (ray-project/ray#65288).
This PR bumps the Ray version in the History Server example YAMLs, e2e test data, and DEVELOPMENT.md from 2.56.0 to 2.58.0.
It also adds the serveLlmSglang dashboard UID to the expected Grafana health response in the e2e test, since Ray 2.58 returns it.

Related issue number

Part of #5356

Checks

• I've made sure the tests are passing.
• Testing Strategy
  • Unit tests
  • Manual tests

Manual test instructions

Ran the History Server e2e tests for /api/v0/tasks (live and dead clusters, including the task log check) and Grafana health on kind with Ray 2.58.0, and they passed.
MinIO was swapped to cgr.dev/chainguard/minio:latest-dev locally because of #5339.

Question

When fixing, found that the time.sleep(2) in test/testdata/rayjob.yaml is a workaround for ray#65288, so it should no longer be needed. Do you want me to remove it in this PR or in a follow-up?

[win5923]

When fixing, found that the time.sleep(2) in test/testdata/rayjob.yaml is a workaround for ray#65288, so it should no longer be needed. Do you want me to remove it in this PR or in a follow-up?

Have you verified that taskLogInfo in the TASK_LIFECYCLE_EVENT is working as expected?

[wuallen57730]

Have you verified that taskLogInfo in the TASK_LIFECYCLE_EVENT is working as expected?

Yes. I checked it two ways, with the time.sleep(2) removed:

1. Raw events from the aggregator agent: on Ray 2.56, taskLogInfo loses stdoutFile and stdoutStart. On Ray 2.58, each task gets one TASK_LIFECYCLE_EVENT with the full taskLogInfo, and the offsets match the printed output.
2. History Server e2e on kind: verifyDeadClusterTaskLogInfo fails on Ray 2.56 as expected, and the live/dead /api/v0/tasks and dead-cluster /v0/logs/file subtests all pass on Ray 2.58.

Outputs:

taskLogInfo for a quick task:

# Ray 2.56
{"stdoutFile": "", "stderrFile": "", "stdoutStart": "0", "stdoutEnd": "49", "stderrStart": "0", "stderrEnd": "36"}
# Ray 2.58 (81 - 68 = 13 bytes = "Processing 0\n")
{"stdoutFile": ".../logs/worker-...-608.out", "stderrFile": ".../logs/worker-...-608.err", "stdoutStart": "68", "stdoutEnd": "81", "stderrStart": "55", "stderrEnd": "55"}

e2e on Ray 2.56:

historyserver_test.go:1947:
    completed Ray 2.56 my_task should expose a complete stdout byte range
    Expected
        <string>:
    not to be empty

e2e on Ray 2.58:

--- PASS: TestHistoryServer/Live_cluster:_/api/v0/tasks?detail=1_should_return_... (178.16s)
--- PASS: TestHistoryServer/Dead_cluster:_/api/v0/tasks_should_return_detailed_task_information_from_historical_replay (157.30s)
--- PASS: TestHistoryServer//v0/logs/file_endpoint_(dead_cluster)/task_id_parameter (0.06s)

So I think the time.sleep(2) can be removed. I can do it in this PR.

[win5923]

Thank you!

[machichima]

kuberay/historyserver/test/e2e/historyserver_test.go

Lines 1947 to 1948 in 9dc5527

	g.Expect(taskID).NotTo(BeEmpty(),
	"completed Ray 2.56 my_task should expose a complete stdout byte range")

nit: also update this to something like "completed my_task should expose a complete stdout byte range"?

[machichima]

Suggested change

	Install Ray locally. Make sure to use at least Ray `v2.58`.

Also update this?

[wuallen57730]

Done, thx!

[machichima]

Thank you!