[mottihoresh]

Part of chunk 5 (#134): items X10, N5 and N8. Base: fix/chunk-5.

X10: one telemetry spec

• Today: the idle placeholders are their own spec. IAS and Press Alt read % FS until the first telemetry tick, then switch to kt and ft. Press Alt's idle id was D_ALT_NEEDLE, not ALT_FT.
• After: there is one spec, TELEMETRY in main/dcs/sim-telemetry-decoder.ts. idleTelemetry() returns the decoder's live readouts with nothing decoded yet, so the gauges show kt, ft and % from the start. The contract's TELEMETRY_IDLE (shared/dcs/ipc.ts) is the renderer's starting value. A test pins it to the decoder's output.
• Side effect: the idle Press Alt ring now uses the secondary tone, the same as the live Press Alt ring. Before, the idle ring was primary and only switched to secondary on the first tick. This follows from having one spec.

N5: host and port

• Today: the host and port are saved on every keystroke, and a port that isn't a number silently saves as 7778.
• After: typing saves nothing. The field is saved on commit (blur or Enter). Main checks the port in settings-form.ts commandPortOf, which accepts whole digits from 1 to 65535. A bad port saves nothing and pushes nothing. config:edit then answers { ok: false, error } (new ConfigEditResult). Both the Settings › Connection form and the Connection tab's DCS Link form show main's error using the existing input--bad / aria-invalid outline and set-err text. No new CSS. A blur that follows an Enter with nothing new typed sends nothing. An import resets the typed text and the error.
• API guard: editConfig now answers ConfigEditResult instead of AppConfig. The frozen copy is updated on purpose, with a header note. No channel was added, removed or renamed.

N8: PSU limit fields

• Today: Settings › Cockpit's custom-limit fields hold their own typed text, which is never refreshed after an import replaces the settings.
• After: an import (and config:get) bumps the config store's formsReset. PowerSupply is keyed on it, so the fields restart from the pushed config. This works the same way as the other form fields, which restart from the config on an import. A form's own edit still leaves its typed text alone.

Tests changed on purpose (they pinned the old behaviour)

• sim-telemetry-decoder.test.ts: the "idle readouts are their own spec" test becomes "one spec": idle equals the live snapshot, with units % RPM / kt / % DN / ft / % QTY. A new test checks that the first frame keeps the idle ids, labels and units.
• t27: the idle gauges read —kt / —ft instead of —% FS.
• settings-edit.test.ts: removed the "junk port saves as 7778" assertions. Added tests for commandPortOf, for a junk port being refused with its error, and for SettingsService.edit saving and pushing nothing on a junk port. Updated the tests for the new SavedEdit / ConfigEditResult shapes.
• pushed-rules.seam.test.ts: config:edit { port: '77a' } is refused with the error. config.json and config:get keep the last good port, and no config:changed is pushed.
• t28 / t31: "saved as typed" becomes "saved on commit". These tests check that typing saves nothing, that blur and Enter each save, and that a blur after Enter does not send again. New tests check that a junk port shows the error with aria-invalid and saves nothing, and that a good port then saves and clears the error.
• t29 (new): renders the Cockpit and Configuration sections together. A typed (refused) limit is followed by an import with other limits, and the fields show the imported limits.
• skyhawk-mock.ts: editConfig answers ConfigEditResult the way main does.

All new and changed tests fail on the old source. I checked this by stashing the non-test changes: 16 tests failed.

How verified

npm run lint, npm run lint:deps, npm run typecheck and npm test (1421 tests) all pass locally, after rebasing onto fix/chunk-5 at 8c48345 (#156). I have not tested on the DCS computer yet; that belongs in the chunk 5 device session.