Run the test suites in CI, and make the fmt and clippy gates pass

[ParallelEntrepreneur]

README.md listed "CI running the Rust, runner and UI test suites" as PLANNED, and said
"No CI yet". Neither was quite true and neither was the whole story: release.yml already
builds, bundles and publishes to crates.io, but nothing ever ran a test. This adds the
workflow that does, and makes the lint gates it enforces actually pass.

What changed

.github/workflows/ci.yml (new). Runs on every pull request, on pushes to main, and on
workflow_dispatch, as three parallel jobs on ubuntu-24.04:

• rust — cargo fmt --all -- --check, then cargo clippy --workspace --all-targets --locked -- -D warnings, then cargo test --workspace --locked.
• node — the Claude Code runner's suite (npm ci + npm test) and the telemetry receiver's
  (node --test; it has no dependencies, so it needs no install step).
• web — npm ci + npm run build (tsc --noEmit && vite build). There is no UI test suite,
  so this is a build check and the README says so.

It follows release.yml's conventions: actions pinned by commit SHA with a # vX.Y.Z comment
(the same pins already in use), rustup toolchain install stable --profile minimal rather than
a setup action, top-level permissions: contents: read, a concurrency group that cancels
superseded PR runs, and timeout-minutes on every job.

A genuinely broken test command, fixed. modules/agents/claude-code's test script was
node --test test/, which fails on Node 24 — the positional argument is treated as a literal
entry file, so it exits 1 in 25ms with Cannot find module '.../claude-code/test' without
running anything. npm test in that package was broken, as was the command documented in
README.md. Bare node --test discovers test/*.test.mjs correctly and passes all 58 tests.
Fixed in the script and the README.

cargo fmt --all across the workspace (26 files, formatting only) and cargo clippy --fix,
so the two new gates go green. The clippy pass touched 10 files: 13 collapsible_if collapses,
one while_let_on_iterator, one manual_contains, one items_after_test_module code motion,
and one hand-written fix for unnecessary_sort_by in memory.rs (sort_by →
sort_by_key(|p| Reverse(...))). No #[allow] was used to silence anything.

rust-version = "1.88" in both crate manifests. The collapsible_if fixes produce
let-chains, which need Rust 1.88 — edition 2024 alone only requires 1.85. Since both crates are
published to crates.io, this is now declared rather than left as an undocumented effective
minimum, so cargo gives consumers a clear error instead of a syntax failure.

README.md — the roadmap row flipped to SHIPPING and reworded to match what CI really does
(three test suites plus a UI build, not a "UI test suite"); the "No CI yet, and nothing is
published to crates.io or npm" bullet rewritten, since only the npm half was still true; and the
Development section now notes CI and shows the npm ci a fresh clone actually needs.

How this was verified

Everything below was run in the sandbox, and then re-run from scratch by a separate agent that
read ci.yml and executed each step's command verbatim:

Gate	Result
cargo fmt --all -- --check	exit 0
cargo clippy --workspace --all-targets --locked -- -D warnings	exit 0, no warnings
cargo test --workspace --locked	100 passed, 0 failed
runner: npm ci + npm test	58 passed, 0 failed
telemetry: node --test	8 passed, 0 failed
web: npm ci + npm run build	exit 0

actionlint v1.7.12 reports the workflow clean. The action SHAs were compared literally against
release.yml. The rustup ... --component rustfmt,clippy line was verified empirically rather
than by eye — clippy was removed from the toolchain and the exact workflow line re-run, which
reinstalled it. The old node --test test/ was run under Node 24 to confirm the failure is real
rather than inferred.

Worth a close look

This branch is based on a stale origin/main and will conflict on merge. It could not be
rebased here (git metadata is read-only in the colony). Simulating the merge in a throwaway
clone: HEAD already conflicts with origin/main on 5 files before any of this work —
Cargo.lock, both crate Cargo.tomls, web/src/components/ChatPanel.tsx and
web/src/sessionStream.ts. The workspace-wide reformat adds 5 more, all Rust: headroom.rs,
main.rs, memory.rs, sessions.rs and telemetry.rs (~28 marker blocks). Most are
style-versus-semantic and resolve by taking main's side and re-running cargo fmt, but
memory.rs needs real care — #97 rewrote 253 lines of it inside a region this reformat also
touched. Note that headroom.rs and telemetry.rs are byte-identical on both sides today; the
reformat alone is what breaks that identity. Rebasing onto origin/main and re-running
cargo fmt --all is likely to be much less work than resolving the reformat by hand.

The two Cargo.toml files that gained rust-version are themselves already in that baseline
conflict set.

The roadmap row is marked SHIPPING, which README.md:78-81 defines as "merged, in this
repository, and exercised on a real machine". That is true by the time this merges — GitHub runs
pull_request workflows from the PR's merge ref, so this PR exercises its own CI — but the
claim is slightly ahead of the evidence until that first run is green. Worth confirming the run
before merging.

The clippy autofixes are the only semantic changes in a very large diff, and each was reviewed
individually: the sort_by_key(Reverse(...)) rewrite preserves ordering and tie-breaking (both
are stable sorts on a Copy key), chars.by_ref() keeps the iterator usable after the loop as
the surrounding code requires, and every collapsed condition with an || was parenthesised to
preserve precedence. The remaining ~3900 lines of Rust diff are pure formatting churn and are
best reviewed with whitespace ignored.

---

🤖 Generated by Colonizer in a microVM

[ParallelEntrepreneur]

Superseded by #142.

This branch's only new commit was a whole-tree cargo fmt plus the gate, on a base that predated eight commits since squashed onto main — which is why it showed 71 changed files and would not merge. The reflow is machine-generated, so #142 redoes it against main instead of rebasing ~4,000 conflicting lines of pure whitespace.

One deliberate difference: #142 adds a rustfmt.toml pinning width 130 rather than taking rustfmt's 100-column default. The tree was written wider (99% of its lines already fit 129 columns), and at 100 the reflow is nearly twice the churn. The CI half of this branch is already on main as #138.