feat(lunar): add a solar/lunar/both mode to the detail view

[daneden]

Stacked on #166 — based on claude/solstice-solar-eclipse-n2k27c so the diff shows only the lunar work. Retarget to main once #166 merges.

Adds a toolbar control that switches the detail view between the sun, the moon, and both.

Why this was cheaper than it looks

EclipseCalculator already held a complete Meeus lunar theory — chapter 47's truncated ELP-2000/82 series, topocentric parallax, nutation, sidereal time and ΔT — and every bit of it was private. None of it is specific to eclipses. The first commit lifts it into Ephemeris; the second builds on it.

The extraction is numerically inert, and checkably so: counting every numeric literal across both files gives 937 before and 937 after. The only differences are the lunation index going from spelled out twice to once, and two new constants. EclipseCalculatorTests is unchanged and is the real check.

One constant deliberately does not become the default. The moon radius in EclipseCalculator was the umbral k = 0.272281, chosen so totality durations match NASA's, and it is the wrong figure for how large the moon looks. Ephemeris exposes the mean radius as moonRadiusKm and keeps the eclipse value as umbralMoonRadiusKm.

The one piece of genuinely new astronomy

Moonrise and moonset. The sun's fall out of a single daily position; the moon moves ~13°/day, so altitude is sampled across the local day and the crossings bisected.

The trap worth knowing about: Meeus gives h₀ = 0.7275·π − 34′ for moonrise, but that is for geocentric positions — the 0.7275·π term exists to fold in parallax. Ephemeris returns topocentric positions, where parallax is already applied, so using Meeus's figure would move moonrise by the better part of two hours. What remains is refraction and the semidiameter, as for the sun. The test that the full moon rises as the sun sets exists to catch exactly that.

Both times are optional, and a missing one is ordinary rather than an error: about once a lunation a calendar day has no moonrise, or no moonset, at every latitude.

Phases

Eight equal 45° slices would label a moon that is 68% lit "first quarter" — a first quarter moon is half lit, and anyone who looked up would see the app was wrong. The four principal phases get a narrow window instead and the crescents and gibbous phases take everything between. Meeus's own worked example falls in that gap and now reads as waxing gibbous.

Glyphs are mirrored below the equator: the moon is lit from the same side for everyone, but southern observers see it the other way up.

Verification

Asserted against the book, which the extraction made possible for the first time — these were previously only checked indirectly, through whether an eclipse landed in the right place:

Reference	Result
Meeus 47.a (lunar position)	λ, β, Δ to six decimal places
Meeus 25.b (solar position)	matches
Meeus 48.a (illuminated fraction)	0.6786

And against physics, where no published example exists:

• The full moon rises as the sun sets — sun altitude at moonrise came out at −0.94°.
• Daily retardation of moonrise is 11–23 min at London in late summer, not the "average" 50 — the shallow-ecliptic-angle effect behind the Harvest Moon.
• Successive full moons a synodic month apart.
• The phase search agrees with the independently written eclipse search about the 2026 new moon, to 11 minutes — the expected offset between conjunction in longitude and greatest local eclipse.

Judgement calls

• The sky gradient stays solar in every mode. It's driven by sun altitude, and keeping it as the backdrop is what makes the moon's path readable — you can see whether the moon is up in darkness or wasted in daylight. It also gives "both" a clean meaning.
• The moon is computed regardless of mode, so switching shows data immediately rather than after a round trip. It costs a few hundred microseconds off the main thread, once per day and place.
• Moon altitude samples use the existing SkyRenderCache pattern rather than .task — the share card renders this chart through ImageRenderer, which never runs the view lifecycle, so anything task-driven would come out blank there.

Not verified

No Swift toolchain in the session, so none of this has been compiled or run — the validation above is via a line-for-line Python port of the same algorithms. Worth a close look on first build:

• The five new SF Symbols (moonrise, moonset, moonphase.*) render as blank space rather than failing the build if a name is wrong.
• SolsticeTests has never executed in CI — both checks on feat(eclipses): surface upcoming solar eclipses and alert on major ones #166 are archive actions.

Follow-up

PR 2 will add the phase calendar replacing the annual chart in lunar mode. Lunar eclipses are deliberately deferred.

---

Generated by Claude Code