docs: review-002 minor fixes — op PATCH semantics, repo-id grammar, ssh tuple, ADR-016 glitches

- backend.md: git/repo/update pinned as PATCH (omitted fields unchanged;
  present fields replaced wholesale) — R-3; 'already_exists' disclosure
  posture recorded (accepted create-scope oracle; resolve-side stays
  collapsed) — R-6; repo-id grammar pinned (owner/name segments, rejection-
  only parsing, percent-encoded flat record-file naming, two-placeholder
  http-route note) — R-5
- doors.md: alkssh hand-off tuple gains service + authorized-repo marker
  (ADR-016/D-3 amendment rounds missed this third tuple site) — R-4
- ADR-016: collapsed-space and broken code-span glitches — R-8
- AGENTS.md: lifecycle paragraph records both gate reviews complete and
  decomposition unblocked

Verification: docs-only; cargo doc/test/clippy/fmt clean

docs/architecture/decisions/016-native-session-preamble.md

@@ -102,7 +102,7 @@ the session tuple gains the service dimension.**
    fail before any advertisement; repo resolution and authorization
    failures collapse per ADR-008 (unknown ≡ unauthorized). This is
    alkgit-specific wire format on a published ALPN — it enters OQ-03's
-   freeze    inventory (one-way). The grammar is also the one a git://-to-
+   freeze inventory (one-way). The grammar is also the one a git://-to-
    `alk/git` bridge would need anyway, and it is the framing
    `GitSession::connect_direct` sends — the two native paths speak one
    preamble dialect.
@@ -120,8 +120,8 @@ the session tuple gains the service dimension.**
      above).
 
 4. **`GitSession` mirrors the same shapes**: `connect_direct` sends the
-   request-line preamble before waiting for the server; `open_via_
-   channels` sends the same `{repo, service}` params schema.
+   request-line preamble before waiting for the server; `open_via_channels`
+   sends the same `{repo, service}` params schema.
 
 5. **Rejected alternatives** (recorded so the decomposer does not
    reinvent them):