v0.4.0 — Honest estimates and polish

A calibration a user can actually carry out, charts that show the reading rather than the axis, and the defects a pass over the sensor lifecycle turned up.

Added

• Calibrate against a reference meter (R-75). Hold a sound level meter or a lux meter beside the phone, type what it reads, and Readout works the correction out itself — averaged over five seconds, rounded no finer than the estimate deserves, and refused when the two figures are too far apart to be a calibration. In the Sound meter, the Light meter and Settings.
• A labelled time axis on the live chart, an empty state for a grid with every tile hidden, a way back to the first-run explainer, and the model identifier a calibration would be filed under.

Changed

• Charts fit the samples on screen instead of reaching down to zero.
• Portrait only, which is what R-1 has always said.
• Tiles grow with Dynamic Type; coordinates sit on two lines.
• In the Simulator every reading comes from the synthetic source, so the location group no longer waits for a fix that never arrives.

Fixed

• A permission granted inside the app now starts whatever was waiting on it.
• The sound reading is no longer hidden on a phone that has a microphone.
• No blocking Core Location check while the first frame is built.
• The spirit level's bubble stays inside its circle on the diagonals.
• The Sound meter's peak comes from the same history as its other statistics.

Built and tested in the Simulator — 127 tests. Nothing here has been verified on physical hardware; the on-device checklist in docs/RELEASE.md §2 still gates v1.0.

Full notes: CHANGELOG.md