UTS: correct stale and inconsistent UTS docs and proxy specs

[sacOO7]

Summary

This PR fixes stale and inconsistent text in the existing UTS docs, and two proxy specs that don't match proxy.md. These problems turned up while writing the translator-skill docs in #558.

The PR is independent of that one and targets main. It changes 4 files (+26 / −5) and adds no new tests or spec points.

Changes

1. Define CLOSE_CLIENT — uts/docs/writing-test-specs.md:661–672

CLOSE_CLIENT(client) appears on 464 lines across 55 spec files. Until now its meaning was recorded only in the message of the commit that added it (3745335). This adds a short "Closing Clients (CLOSE_CLIENT)" subsection under Pseudocode Conventions. It says that:

• the call closes the client so its internal timers stop;
• it is a cleanup step, not an assertion, so it doesn't wait for or assert CLOSED;
• derived tests should close the client even when the test fails, pointing to "Cleanup with afterEach" in writing-derived-tests.md.

Other undocumented pseudocode constructs (AWAIT_ERROR, AWAIT UNTIL, IS SAME AS, …) are deliberately left out. Picking canonical spellings for them is for the spec owner to decide.

2. Add the objects Test ID categories — uts/docs/writing-test-specs.md:38

The category field said "One of:" followed by the rest and realtime categories only. The corpus has 339 Test IDs under objects/unit, objects/integration and objects/proxy, which follow the same <module>/<tier> convention. This adds those three categories to the list.

3. Use endpoint: in the proxy test template — uts/docs/writing-test-specs.md:346

The "Proxy Test Structure" template called create_proxy_session(target: TargetConfig(realtimeHost: …, restHost: …), …). That conflicts with the signature in proxy.md (create_proxy_session(endpoint: String, …)) and with every call in the corpus. The template now uses endpoint: "nonprod:sandbox", so spec authors who copy it get the documented form.

4. List objects_gc_test.md under Protocol Variants — uts/docs/integration-testing.md:305

objects/integration/objects_gc_test.md has a ## Protocol Variants section (added in 61b5afd), but the "Annotated specs" list didn't include it. The list now matches all 13 annotated specs.

5. Pass endpoint to create_proxy_session — uts/realtime/integration/proxy/presence_reentry.md:74–77, 197

proxy.md declares endpoint as a required argument, but both calls in this spec passed only rules:, which left each SDK harness to invent a default. Both now pass endpoint: "nonprod:sandbox", formatted like the other calls. All 43 create_proxy_session calls in the corpus now pass the same endpoint.

6. Replace __PASSTHROUGH__ with a fixed fake key — uts/realtime/integration/proxy/connection_resume.md:782–785, 796, 798 (Test 22, RTN14h; separate commit)

The test's replace rule set connectionKey (top level and in connectionDetails) to "__PASSTHROUGH__". Neither proxy.md nor uts-proxy defines any such placeholder. uts-proxy's replace action sends the message verbatim (ws_proxy.go, case "replace": dst.WriteMessage(…, rule.Action.Message), at v0.3.0 and on main), so the client has always received that literal string. The name implied that the proxy preserves the server's real key, which could lead harness authors to build a substitution that doesn't exist.

The key is now "proxy-ttl-test-key", in the style of the file's other fake keys (proxy-injected-new-key, recovery-failed-new-key), with a comment explaining that the message is sent verbatim. Because the client always received a literal value, its behaviour is unchanged, and none of the test's assertions depend on the key.

Needs acknowledgement

Item 6 (Test 22 in connection_resume.md) needs an acknowledgement from the test's author. It assumes the intent was "any key; the resume is expected to fail server-side", as the test's prose and its assertion connection.id != "proxy-ttl-test-id" suggest. If the intent was for the client to resume with the server's real key, uts-proxy would need a field-preserving replace, documented in proxy.md, instead of this change. Item 6 is its own commit, so it can be dropped without affecting the rest.

Not in this PR

• The translator-skill docs themselves: UTS: add translator-skill guide and uts-to-lang-skill-creator skill #558. That PR's description lists these follow-ups. Some guide text in it (the notes on rules:-only sessions, target: and __PASSTHROUGH__) can be simplified once this PR merges.
• A separate follow-up PR, stacked on PR 558, will align three existing rules with the guide so its "Divergence from …" labels can go:
  • AWAIT_STATE subscribes, then checks;
  • work is drained before negative assertions;
  • the features spec is read at the same revision as the UTS spec.
• A GitHub issue will raise two decisions for the spec owner:
  • which deviation category covers SDK APIs that are missing (spec-correct tests that can't compile);
  • canonical spellings for the other undocumented pseudocode constructs.

How verified

• Each claim was rechecked against current main with git grep: 464 CLOSE_CLIENT lines, 339 objects Test IDs, 13 ## Protocol Variants files, and 43 of 43 create_proxy_session calls passing endpoint: "nonprod:sandbox" after the change. The uts-proxy replace behaviour was read at tag v0.3.0 and on main.
• git diff --check is clean. editorconfig-checker shows no new errors on the changed files: the 3 indentation warnings in writing-test-specs.md exist on main and moved down 12 lines. The repo lint (build/scripts/find-duplicate-spec-items) passes. The changed lines add no Markdown links, so no links or anchors are affected.
• An independent read-only reviewer re-derived every claim from the sources and the uts-proxy code, and confirmed each change is correct and minimal, with no out-of-scope edits.