Unselected Settings tab names are white at 56% on the rail, 4.15:1

## What happens

In the Settings window's rail, a tab that isn't selected draws its name in white at 56% opacity (`Sources/Uttrflow/Settings/SettingsRootView.swift:97`, `.foregroundStyle(.white.opacity(isSelected ? 1 : 0.56))`), 12.5 pt medium.

The tabs sit at the top of the rail, where the gradient is `BrandPalette.Teal.railTop` `0x0E4F49`. White at 56% over that colour is `0x93AFAC`, and its contrast is **4.15:1**, under the 4.5:1 WCAG AA asks of text at this size. It's the same in both appearances, because the rail doesn't change with them.

| Opacity | Contrast on `railTop` |
|---|---|
| 0.56 (now) | 4.15 |
| 0.60 | 4.52 |
| 0.62 | 4.68 |
| 0.66 | 5.08 |

## Why it matters

Four of the five tab names are always the unselected ones, so most of the Settings navigation is below AA for low-vision users.

## How to reproduce

Open Settings (⌘,). Compute the ratio from the values above with the WCAG 2.x formula, or measure an unselected tab name against the rail with Digital Color Meter.

## Acceptance criteria

- Unselected tab names reach at least 4.5:1 against `railTop`. 0.62 gives 4.68:1 and still reads clearly as unselected next to the selected tab's full white and semibold weight.
- Nothing else about the rail changes.

## Where to start

- `Sources/Uttrflow/Settings/SettingsRootView.swift:97`. Consider naming the opacity in `SettingsMetrics` with a comment giving the ratio.
- Palette tests live in `Tests/UttrflowTests/OrbitPaletteTests.swift` if you want to pin the ratio.
- Run `make verify`, and see `CONTRIBUTING.md`.

Size: one line.

Start at Sources/Uttrflow/Settings/SettingsRootView.swift:97 and inspect the unselected tab foreground opacity. Adjust only that opacity so the tab names reach at least 4.5:1 contrast against BrandPalette.Teal.railTop, then run make verify. Review Tests/UttrflowTests/OrbitPaletteTests.swift if adding a contrast check; done means the rail is otherwise unchanged and the acceptance ratio is met.

Written by the indexing model from the issue text.