refactor(stella-time): one home for the real time sources behind the Sleeper and Clock ports - #6488

## What & why

`main` is red on `lockfile-sync` and every `--locked` build (the
canary's report is #6495). The `stella-time` crate landed in #6488 with
its `Cargo.lock` entry at `0.9.416`, while the release sync in #6494 had
already moved every workspace crate to `0.9.417`. Neither tree was wrong
on its own; the composition is. This regenerates the lock, which changes
exactly that one line.

Refs #6495

## The witness

- [x] No witness needed (lockfile only) — because: `cargo metadata
--format-version 1 --locked` fails on `main`'s tip and passes with this
lock. That command is the `lockfile-sync` gate step.

## The gate

- [x] `make guards-fast` (lockfile-sync included) passes locally
- [ ] the rest is CI's

## Fix over file

- [x] Extra fixes in this PR: none
- [x] Nothing was deferred

## Anything reviewers should know?

Carries `unblocks-main` so the red-main hold does not block its own
repair, and `no-issue` because the canary's own run marks that issue
resolved once `main` recovers; nothing here closes it.

…rp from the docs site (#6498)

## What & why

Two follow-ups to #6488, each its own commit.

- **`make core-no-io` told test code from shipping code by file name
alone** (a `tests/` directory or a `tests.rs` file). A module declared
under `#[cfg(test)]` by any other name was read as shipping code, which
is why #6488's shared sleeper doubles had to be named `src/tests.rs`.
The guard now reads every `#[cfg(test)] mod name;` line and skips the
file or directory it names, at any depth. `scripts/test-core-no-io.sh`
gains the case and its control (a plain `mod` line must still count), 29
cases in all.
- **`sharp` is removed from the docs site instead of floored.** The site
imports no `next/image`, so sharp was never loaded: inert weight from
`next`'s optional dependencies that carried the advisory #6488 floored
and fourteen LGPL-3.0 libvips tuples that `dependency-review` had to be
told about by name on every `next` bump (the recurrence issue #2532
describes). The pnpm override drops it from the tree, its build-script
approval goes with it, and the per-tuple exemption #6488 added to
`dependency-review.yml` is deleted, since nothing is left to exempt. The
`js-yaml` floor stays.

Refs #6484. Closes nothing by design (`closes-nothing`).

## The witness

- [x] This PR includes a witness test (fails on `main`, passes here)

`scripts/test-core-no-io.sh`'s new case `a module named under
#[cfg(test)] is not shipping code, by any name and at any depth` fails
against `main`'s guard (the file it names reads `std::fs`) and passes
here; its control `a module named by a plain mod line is shipping code`
pins that the rule did not widen. For the sharp change the evidence is
the lockfile: `sharp` appears once in `website/pnpm-lock.yaml`, in the
override line, and `docs.yml`'s `typecheck + build` job is what proves
`next build` does not need it.

## The gate

- [x] `cargo fmt --check` (no Rust touched)
- [x] clippy — no Rust touched
- [x] tests — the guard harness ran locally, 29 passed; `make prose`,
`make license-allowlist-parity` and `make action-pins` green
- [x] Docs updated where behavior changed (the guard's own docstring and
AGENTS.md's no-io paragraph name the new rule)
- [x] CLA signed

## Fix over file

- [x] Extra fixes in this PR: none beyond the two named
- [x] Nothing was deferred

## Ground-rule check

- [x] No I/O added to `stella-core` (not touched)
- [x] No new outbound network calls
- [x] No new cross-boundary serde types

## Anything reviewers should know?

- If the docs site ever renders a `next/image`, restore a `sharp` floor
in `website/pnpm-workspace.yaml` and put the tuples back in the
workflow; the comment there says so.
- The stacked doubles PR (#6491) was closed when #6488's branch was
deleted on merge; it is re-opened against `main` separately and does not
depend on this one.

## Summary by Sourcery

Correct test-code classification in the core no-I/O check and remove
unused Sharp dependencies from the documentation site.

Bug Fixes:
- Update the core no-I/O guard to exclude modules declared under
`#[cfg(test)]` regardless of their file name or nesting depth, while
preserving shipping-code detection for plain modules.

Enhancements:
- Remove the unused Sharp dependency from the documentation site and
retain the existing js-yaml security floor.

CI:
- Remove Sharp and its platform-specific license exemptions from
dependency-review configuration and its build-script approval.

Documentation:
- Document the expanded test-module exclusion rule in the core no-I/O
guard guidance.

Tests:
- Add regression and control cases covering cfg-gated modules at
multiple depths and plain modules.

…a-time (#6499)

## What & why

The test-double half of #6484, on top of the merged `stella-time` PR
(#6488). Re-opened: the first copy (#6491) was closed by GitHub when
#6488's branch was deleted on merge. About thirty test files across
`stella-core`, `stella-engine` and their `tests/` directories each wrote
the same `Sleeper` double — a no-op one under three names, and one on
tokio's paused clock — and each copy grew a `now()` body in #6486. Every
one of them is deleted here.

- Integration tests and `stella-engine` take `PausedSleeper` and
`NoopSleeper` from `stella_time::test_util` through a dev-dependency on
`stella-time` with `test-util` on. For `stella-core` that is a
dependency cycle, which cargo allows; it is the tokio / tokio-test
shape.
- `stella-core`'s own unit tests keep one copy, `src/tests.rs`, because
a lib's unit tests are a second build of the lib and a dev-dependency
that links the lib implements the trait for the first. The compiler
forces that copy (the first attempt at taking the crate's doubles from
`stella-time` failed on every unit test with "multiple different
versions of crate `stella_core`"). It is named `tests.rs` so `make
core-no-io` reads it as test code, which is how that guard tells test
code from shipping code.
- The `one_home` witness in `stella-time` now scans test files too. Its
kept list shrinks to the four doubles with a shape a shared one cannot
have — the retry tests' recording sleeper, the monitor's advancing
sleeper, the hanging sleeper, and `step/tests.rs`'s unpaused-time
sleeper — plus the forced copy.

Closes #6484

## The witness

- [x] This PR includes a witness test (fails on `main`, passes here)

`crates/stella-time/tests/one_home.rs` reads every `.rs` under
`crates/*/src` and `crates/*/tests`. On the base branch it fails on
twenty-three files that each define a sleeper double; here it passes,
and a second test fails if a kept one disappears. Locally:
`stella-core`'s 893 lib tests and every integration test,
`stella-engine`'s tests and the time crate's 6 + 3 tests pass.

## The gate

- [x] `cargo fmt --check`
- [ ] clippy over the workspace at `-D warnings` — CI; ran locally on
`stella-core`, `stella-engine`, `stella-time`
- [ ] the workspace test suite — CI
- [x] Docs updated where behavior/flags changed (README, `--help`, doc
comments)
- [x] CLA signed
- [x] `Closes #N` appears **both** above and as a commit trailer

## Fix over file

- [x] Extra fixes in this PR: none beyond the sweep
- [x] Nothing was deferred

## Ground-rule check

- [x] No I/O added to `stella-core`; `src/tests.rs` is `#[cfg(test)]`
and `make core-no-io` is green
- [x] No new outbound network calls
- [x] No new cross-boundary serde types

## Anything reviewers should know?

- The rename `TokioSleeper` → `PausedSleeper` in the test tree is
deliberate: the paused-clock double must not share a name with the real
`stella_time::TokioSleeper`.
- Once #6498 lands (the no-io guard reads `#[cfg(test)] mod` lines),
`src/tests.rs` could take a more descriptive name; it is named for the
guard's current file-name rule.

## Summary by Sourcery

Consolidate test sleeper doubles in `stella-time` and enforce a single
ownership location across the workspace.

Enhancements:
- Centralize the shared no-op and paused-clock sleeper doubles in
`stella-time` and update `stella-core` and `stella-engine` tests to use
them.
- Retain only the specialized sleeper doubles that require unique
behavior, including the compiler-required `stella-core` unit-test copy.
- Extend the sleeper ownership witness to scan test sources and document
the shared test utility arrangement.

Build:
- Add `stella-time` as a test-only dependency with the `test-util`
feature where shared sleeper doubles are used.

Documentation:
- Update architecture and crate documentation to describe the shared
sleeper test utilities and the `stella-core` unit-test exception.

Tests:
- Update the affected unit and integration tests to use the centralized
sleeper doubles.