fix(features): 保留键的诊断指向一个存在的拼写

自审读出:`deps` 的那条消息提供了 `optional = true`,而 mcpp 从来没有这个键。
一条把读者送去一个解析器不认识的键的诊断,与本次发布正在移除的那些警告是同一个
缺陷,只是外了一层。文档化的机制是 `[feature-deps.<name>]`(docs/05 §2.8.2)。

同步中英两份 docs/05。

Author: speak-agent

.agents/docs/2026-08-31-issue540-seven-audit-findings.md

@@ -532,6 +532,13 @@ pass that always fires both produce a green single-leg test.
 
 ### 6.9 D16 — the third option: finish the channel that already exists
 
+> **Not taken.** D9 shipped instead, for the cost-floor reason argued at the end
+> of this section: a package wanting one conditional flag should not have to
+> ship and run a C++ program. **D16a was taken anyway**, because it is a defect
+> rather than an option — `docs/05` described `target_libc()` as the resolved C
+> library when it names the payload, and that misleads authors under any of the
+> three answers. D16b and D16c were not needed once the declarative axis worked.
+
 Do not build a predicate axis. Close the three gaps in §6.7 instead:
 
 - **D16a.** Make the resolved `c-abi` reachable from a build program — either by
@@ -689,7 +696,7 @@ The #531 path does neither. §12 is what that costs.
 | 5 | D1 + D3 + D2a + D4 — the text corrections and the `[features]` check | mechanical; no interaction with anything above |
 | 6 | D11 — `docs/13` and `docs/17`, four files | after 2, so the documents describe the gated behaviour |
 | 7 | **D16a** — `target_libc()` reports the payload ref while `docs/05` calls it the resolved C library (§6.7 item 2) | a defect on its own terms, required under all three options below, and it misleads package authors today |
-| 8 | D9 / D16 / D10 — the declarative axis, the imperative channel, or withdrawal | the only item needing a decision rather than an implementation |
+| 8 | D9 / D16 / D10 — the declarative axis, the imperative channel, or withdrawal | the only item needing a decision rather than an implementation. **Decided: D9**, plus D16a because it is a defect either way (§6.9) |
 | 9 | D15 — the tests §12.4 names | after the behaviour they assert exists |
 
 Items 1 through 7 are unambiguous. Item 8 wants a maintainer's answer, and D8

docs/05-mcpp-toml.md

@@ -1331,8 +1331,8 @@ simd       = { sources = ["src/simd/**"], flags = [
 
 - **The table form accepts exactly** `implies`, `forward`, `defines`, `sources`,
   `flags`, `requires`, `provides`. Anything else is reported as a schema warning
-  and ignored (mcpp 2026.9.1.1+); `deps` is reported separately as reserved,
-  because it is planned rather than wrong. Before that release `[features]` was
+  and ignored (mcpp 2026.9.1.1+); `deps` is reported separately as reserved and
+  points at `[feature-deps.<name>]`. Before that release `[features]` was
   the one structured section with no schema check at all, so a misplaced
   `include_dirs` inside a feature built successfully with no diagnostic while
   the identical mistake in `[build]` was reported.

docs/zh/05-mcpp-toml.md

@@ -1150,7 +1150,7 @@ simd       = { sources = ["src/simd/**"], flags = [
 
 - **表形式恰好接受** `implies`、`forward`、`defines`、`sources`、`flags`、
   `requires`、`provides`。其余键会被报成一条 schema 警告并忽略(mcpp 2026.9.1.1+);
-  `deps` 单独报为「保留」,因为它是计划中的而不是写错的。在该版本之前,`[features]`
+  `deps` 单独报为「保留」(它是计划中的而不是写错的),并指向 `[feature-deps.<name>]`。在该版本之前,`[features]`
   是唯一一个完全没有 schema 检查的结构化段落 —— 把 `include_dirs` 误写进 feature 里
   会零诊断地构建成功,而同样的错误写在 `[build]` 里会被报出来。
 - `defines` 为**裸**宏名(不带 `-D`);feature 激活时每个脱糖为 `-D<x>`,加到该包

modules/manifest/src/toml.cppm

@@ -530,11 +530,18 @@ std::expected<Manifest, ManifestError> parse_string(std::string_view content,
                     // "unsupported" would deny a documented plan; saying nothing
                     // is what let it look implemented.
                     if (fkey == "deps") {
+                        // ⚠️ THE SPELLING NAMED HERE HAS TO EXIST. The first
+                        // draft of this message offered `optional = true`,
+                        // which mcpp has never had — a diagnostic that sends
+                        // its reader to a key the parser does not know is the
+                        // same defect as the warnings this release removes,
+                        // just one layer out. `[feature-deps.<name>]` is the
+                        // documented mechanism (docs/05 §2.8.2).
                         m.schemaWarnings.push_back(std::format(
-                            "[features].{}.deps is reserved for a later stage and "
-                            "is not read yet (ignored). Declare dependencies in "
-                            "[dependencies] and gate them with "
-                            "`optional = true` + `implies`.", fname));
+                            "[features].{}.deps is reserved for a later stage "
+                            "and is not read yet (ignored). To pull in a "
+                            "dependency when this feature is active, declare it "
+                            "under [feature-deps.{}].", fname, fname));
                         continue;
                     }
                     std::string supported;