docs: 更正三处无实测支撑的性能断言

用户指出现有文档里的性能数字可疑。逐个实测后发现三类问题，都不加改原文地
标注更正（历史条目保留原文，仓内已有此惯例）。

## ① 「崩溃到恢复 <1s」——从未成立

写于 v1.0.0 发版说明。但**当时的退避代码就已是 1s**（查 v1.0.0 tag 的
`procRestartBackoff = time.Second`），首次重启就要等 1s。

实测（新增临时测试测量 scheduleProcRestart 延迟）：

    第 1 次崩溃 → 1s      第 2 次 → 2.001s      第 3 次 → 3.002s

顺带纠正我自己刚在站点写错的阈值：并非「崩 3 次停下」。实测第 **4** 次
才停（`procMaxRestarts=3`，判定为 `n > 3`），前 3 次都会重启。

## ② 「RPC 往返 p50 24.1µs」——量级对、数字不符

实测 `BenchmarkToolInvoke`：inline/small **30.4µs**、frame/small 51.5µs、
inline/large 767µs、frame/large 398µs。原文与实测同为几十微秒量级，
但具体值对不上，且未注明测的是哪种 payload。

## ③ 「CLIP 实测常驻 1.15GB」——采样点不对（6 处）

实测加载 chineseclip 两塔，RSS 会**自己降下来**：

    加载前      0.00 GB
    两塔加载后  1.59 GB   ← 峰值
    GC + 静置     0.89 GB   ← 稳态（内核回收未用页）

1.15GB 落在两者之间，既不代表峰值也不代表稳态。线上稳态实测 0.39~0.58GB
（更长时间静置后更低）。同源问题：qwen3vl 的「常驻 9.4GB」实为**峰值**，
其视觉塔本就是按需加载（源码注释：每张图约 1.6GB，故按需）。

6 处全部改为「稳态 X（峰值 Y）」双值，消除口径歧义：README 中英、
docs/zh/multimodal-space.md、config/registry.go（2 处 + 1 处注释）、
providers/chineseclip/tokenizer.go。

## 验证

- `go build`（含 `-tags onnxruntime` 与不带）与 `go vet` 均通过
- 全仓 `grep 1.15GB` 已清零
- 测量用的临时测试文件已删除，无残留

README_EN.md

@@ -236,7 +236,7 @@ External plugin development: see [homeagent-sdk](https://gitcode.com/JianFeeeee/
   It exposes only a public provider SPI (`pkg/embedding`: `Modality` / `Input{Data,MIME}` /
   `Info{Dimension,Fingerprint,Modalities}` + a name registry), with implementations under
   `providers/*`. Default: **Chinese-CLIP ViT-B/16** — text and image land in the **same space**
-  (512-dim, fingerprint `cd2a495cf990`, Apache-2.0, ~1.15GB RSS measured standalone);
+  (512-dim, fingerprint `cd2a495cf990`, Apache-2.0; measured ~1.59GB peak on load, settling to ~0.89GB steady-state);
   `qwen3vl` is kept (2048-dim, ~9.4GB) for machines with headroom or future video. Text search
   still falls back to the existing word-vector / TF-IDF path — a CLIP dual tower's pure-text
   semantics are **weaker than an MLLM-style embedder**, a cost documented rather than hidden.
@@ -263,6 +263,17 @@ External plugin development: see [homeagent-sdk](https://gitcode.com/JianFeeeee/
 
 > The historical entries below are kept verbatim to show the evolution; two mechanisms in them
 > were **removed in v1.2.0**: text-description-based media indexing, and reference-counted media GC.
+>
+> **Two performance claims also need correcting** (measured 2026-09-20):
+> "crash-to-recovery under 1s" does not hold — restarts are **linearly backed off**, i.e.
+> 1s / 2s / 3s (`procRestartBackoff=1s × nth crash`). Even the *first* restart waits 1s.
+> And the **4th** crash within a 5-minute window stops automatic restarts pending human
+> intervention (`procMaxRestarts=3`, tested as `n > 3`).
+> The backoff was already 1s when this claim was written (v1.0.0), so it never held.
+> "RPC round-trip p50 24.1µs" is the right order of magnitude but does not match current
+> measurements: `BenchmarkToolInvoke` on this machine gives inline/small **30.4µs**,
+> frame/small 51.5µs, inline/large 767µs, frame/large 398µs.
+> The original text is left unedited rather than rewritten, so the history isn't falsified.
 
 **v1.1.1** — Multimodal reaches the **plugin boundary**. v1.1.0 gave the memory system binary
 multimedia nodes, but that path was open only to the kernel itself; this release opens it to