feat(rate-of-closure, #4103): port the flight integrator (the last blocking symbol)

[dieterolson]

The last and hardest of the ten symbols that gated the React migration. I deferred this twice — in #4533 and #4541 — because main's in-file RK4 loop carries #4518's ground-crossing guard, and a naive swap would revert it.

What changes

• flight.ts gains AngularFlightPoint, FlightSimulationOptions, simulateFlightWithOptions
• FlightResult.trajectory widens from FlightPoint[] to AngularFlightPoint[]
• The in-file RK4 loop moves to a new flightIntegrator.ts; both entry points delegate to it, so there is one integrator rather than two that can drift
• Nine bindings the moved loop was the only user of are dropped from flight.ts

Why the swap does not revert #4518

main's guard:

if (pos[2] >= 0.0 && nextPos[2] < 0.0 && t > dt) {

The && t > dt was added because a descending launch starts below zero, tGround can land before zero, and a negative-time trajectory point makes the metric contract raise RangeError — surfacing a bad launch as an error rather than the nonconverged result it is.

The integrator's contact test:

if (currentGap <= 0 || nextGap > 0) return null;

with terminalGapM defaulting to position[2]. It requires currentGap > 0 strictly, so a launch starting at ground level yields a gap of zero and never records a crossing. That is structurally the same protection, expressed as a state condition rather than a time check.

This is not argued from the code alone. All 1,420 tests on main pass unchanged with the integrator swapped in — including wind.test.ts (the Python/TS parity tolerance from #4513) and the ball_flight_metrics_golden_v1 and inverse_flight_solver_golden_v1 Python-parity fixtures. Those goldens are the strongest available evidence that trajectory output is unchanged.

One new bound, no behaviour change

The integrator checks Math.ceil(maxTime / step) > MAX_FLIGHT_INTEGRATION_STEPS (50,000) upfront, keeping synchronous UI-thread RK4 work finite. The default 10 s at a 1 ms step is 10,000 steps, so no existing caller is affected.

What it unlocks

flightGroundTransfer (+260) and simulationTypes (+83) both need the angular state at landing that AngularFlightPoint carries. Their 15 tests land with them.

Two main-owned tests restored (again)

TorqueProfilePanel.test.tsx and PrimaryViewTabs.test.tsx are reverted to main's versions. Both are purely additive by line count yet fail against main's unchanged components — the recurring trap in this migration, and the reason the prune script now refuses to delete files present on origin/main.

Verification

• tsc --noEmit — clean
• eslint . — clean
• vitest run — 1,435 passed across 176 files, 0 failed
• npm run build — production Vite build succeeds
• Staged diff: 6 files, zero deletions outside the integrator move
• SPEC.md: exactly one insertion (row 1.17.36)

🤖 Generated with Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.