[Klaud Cold] docs: add TPUv7x Ironwood Ghostfish as Coming Soon SKU / 文档：新增 TPUv7x Ironwood Ghostfish（即将支持）

[functionstackx]

Summary

• Adds TPUv7x Ironwood Ghostfish with status Coming Soon 🔜 to the supported-hardware table in README.md and README_zh.md, inserted between H100 ✅ and MI455 UALoE72 Coming Soon 🔜.
• Docs-only change; head commit carries [skip-sweep] so no benchmark setup runs.

中文说明

• 在 README.md 与 README_zh.md 的硬件支持表中新增 TPUv7x Ironwood Ghostfish（Coming Soon 🔜），位于 H100 与 MI455 UALoE72 之间。
• 仅文档改动；head commit 含 [skip-sweep]，不触发基准测试。

🤖 Generated with Claude Code

[claude]

Deferring for a human to confirm the new SKU name — not a bug, just a naming-convention check the bot can't resolve on its own.

Extended reasoning...

Overview

This is a two-line, docs-only change adding a row to the hardware-support table in README.md and README_zh.md: "TPUv7x Ironwood Ghostfish | Coming Soon", inserted between H100 and MI455 UALoE72. No code, config, or benchmark logic is touched, and the commit carries [skip-sweep].

Security risks

None. This is a static markdown table edit with no executable content, links, or user input.

Level of scrutiny

Mechanically this is as simple as it gets — a single markdown row duplicated across two files. However, I noticed the new entry breaks the established convention in this table: every other row is either a shipped SKU with a real, verifiable vendor name (✅) or an anonymized placeholder for an unannounced/embargoed chip ("Chip #1 from Hardware Vendor #1", etc.). "TPUv7x Ironwood Ghostfish" is unusual in that it mixes a real Google TPU generation/codename (TPU v7 "Ironwood") with an unfamiliar suffix ("Ghostfish") that doesn't correspond to any TPU codename I'm aware of, and Google TPUs are not typically third-party benchmarkable hardware in the way the other listed SKUs (GPUs from NVIDIA/AMD) are. The bug-hunting pass flagged this exact SKU as possibly fabricated and a verifier ruled it out, but I don't have an independent way to confirm the name is accurate roadmap information rather than a hallucinated/placeholder name that slipped past the anonymization convention used elsewhere in the table.

Other factors

Given the change is trivial to revert and the risk is purely reputational/accuracy (a public roadmap claim), I'd rather have a human with actual knowledge of upcoming hardware partnerships confirm the SKU name before merge, rather than rubber-stamp a specific, brand-named claim about unreleased hardware.