docs: document the WallClockInstant alias exception in code-style rules

[sanity]

Problem

PR #3871 (merged as ac761dab) introduced a use std::time::Instant as WallClockInstant; alias exception in crates/core/src/ring.rs::connection_maintenance to satisfy the suspend/resume detector's need for a real wall clock (comparing CLOCK_BOOTTIME against CLOCK_MONOTONIC, which TimeSource cannot express in tests). The Claude Rule Review agent on that PR correctly flagged that the rule documentation in .claude/rules/code-style.md still reads "DO NOT use: std::time::Instant::now()" with no carve-outs — leaving a future reviewer with no rule to cite as justification for the existing WallClockInstant::now() call sites.

AGENTS.md rule says documentation updates must land in the same PR as the pattern they describe. The doc commit was ready on the fix/nightly-sim-cm-race branch locally, but by the time the rule-review agent flagged it, the merge queue had already frozen the branch (GH006: Protected branch update failed ... A pull request for this branch has been added to a merge queue), so it lands here as a fast-follow instead.

Solution

Add a dedicated "Exception: real wall-clock comparison against boot_time::Instant" subsection under the existing WHEN you need time/rng/sockets in crates/core/ rule. The exception:

• Names the one legitimate use case (OS suspend/resume detection).
• Explains why TimeSource is the wrong abstraction for this specific case (it returns simulation time in tests and can't reflect a real suspend).
• Documents the exact mechanics: alias at the use line with a load-bearing comment, call via WallClockInstant::now(), keep clock samples back-to-back with no intervening .await.
• Points at crates/core/src/ring.rs::connection_maintenance and classify_suspend_jump as the canonical example.
• Explicitly narrows the exception: do NOT add new call sites for any other purpose — every other "I need real time" urge in crates/core/ should still go through TimeSource.

A future reviewer seeing WallClockInstant::now() now has a rule to cite; a future contributor tempted to reach for std::time::Instant::now() under a different justification will see the rule refuses it.

Testing

Docs-only change. No code paths modified. cargo fmt/clippy/test not re-run since nothing Rust changed.

Follow-up to

#3871 (merged as ac761dab).

[AI-assisted - Claude]

[github-actions]

Rule Review: No issues found

Rules checked: git-workflow.md, code-style.md
Files reviewed: 1 (.claude/rules/code-style.md)

The PR adds documentation for the WallClockInstant alias exception. I verified:

• Commit subject follows conventional commits (docs:) and is under 72 characters
• The canonical example referenced (crates/core/src/ring.rs — WallClockInstant, connection_maintenance, classify_suspend_jump) all exist in the codebase at the cited locations
• The documented procedure (alias at use line, load-bearing comment, WallClockInstant::now() at call sites, back-to-back clock samples) accurately describes the pattern already in production code
• No code changes — docs only, so no .unwrap(), spawn, retry, or test-coverage checks apply
• This is a docs: PR, not a fix:, so the regression-test requirement does not apply

No rule violations detected.

---

Rule review against .claude/rules/. WARNING findings block merge.