First PR of the DST hardening work (spec 05-spec-dst, staged into separate PRs). This one locks in the behavior we already have with an exhaustive guarantee matrix, and makes the test environment honest.

What's here

• DST guarantee matrix (src/time/dst.test.ts, 30 cases) covering §3.1 spring-forward, §3.2 fall-back, §3.3 negative controls, §3.4 long stress, across the spec's reference zones: New York, Lord Howe (30-min DST), Chatham (45-min offset), Havana (midnight DST), Auckland (southern hemisphere), London, Tokyo, UTC.
• Sequences are asserted against the spec's invariants, not magic numbers: strictly monotonic in absolute time, consecutive gap >= the expression interval (no rapid-fire), and every returned instant satisfies the expression — which by construction proves no non-existent spring-forward wall-clock ever fires.
• CI runs in TZ=UTC, plus a dedicated Timezone independence job that runs the whole suite under TZ=America/Sao_Paulo to prove a task with an explicit timezone is unaffected by the system timezone.

Note

The current matcher (after the recent MatcherWalker/localTimeToTimestamp rewrite) already passes all 30 scenarios — including 30-min and 45-min zones, midnight gaps, and 1000-iteration per-second/minute monotonicity. This PR is characterization: it proves and pins the guarantees so future changes can't regress them.

Skipped (depend on features that don't exist yet)

• FI-6 startAt/stopAt, FI-7 getNextRuns(n), FI-2 LockProvider — out of scope for DST.

Next PRs

• PR2: utcOffset (fixed numeric offset, no DST) + its scenarios (CN-3).
• PR3: fuzz testing (§3.6).
• DST model docs (§1 → nodecron.com) — separate, in the site repo.

Note: touches .github/workflows/, so merge via the GitHub UI.

Full suite green: 310 tests under both TZ=UTC and TZ=America/Sao_Paulo.