[CI] Shard long kernel test groups

[khluu]

Why

In the 24-hour CI dashboard window ending 2026-09-01 10:31 UTC, these kernel jobs exceeded 45 minutes at P90:

• AMD Attention Kernels shard 1: 3,877s (17 runs)
• AMD Attention Kernels shard 2: 3,044s (17 runs)
• NVIDIA B200 Kernels: 2,994s (8 runs)
• NVIDIA Quantization Kernels shard 1: 2,821s (11 runs)

What changed

• expand Attention Kernels from 2 to 7 pytest shards; the AMD mirror deliberately inherits the same fan-out because both existing AMD shards are over the threshold
• expand NVIDIA Quantization Kernels from 2 to 6 pytest shards; retain the AMD mirror's two-shard declaration for when ci-infra indexSelectLargeIndex: block: [308,0,0], thread: [95,0,0] Assertion srcIndex < srcSelectDimSize failed. #473 starts honoring mirror-local parallelism
• split the B200 command list into three timing-balanced static buckets with fail-closed shard dispatch
• set 30-minute shard timeouts and shard-qualified labels

This consolidates and replaces draft PRs #52340, #52341, and #52349. No merged or active non-draft PR covers these jobs.

Validation

• YAML parse
• current ci-infra Buildkite step schema
• exact B200 command coverage (every original command appears once)
• POSIX shell syntax for the B200 dispatcher
• git diff --check

Targeted CI

Build #86579 passed all targeted jobs. Attention shards: 6:43, 6:48, 6:58, 7:19, 7:35, 7:40, 7:45; B200 shards: 15:24, 16:35, 17:17; Quantization shards: 7:20, 7:25, 8:49, 8:55, 10:46, 13:34.

This is a draft pending human review. AI assistance was used.

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[yewentao256]

LGTM, thanks for the work!