Composite stereo TX recording — true on-air timeline (WAV + API/UI)

[M0LTE]

What

Adds a composite recording: capture a chosen set of transmitters into a single, sample-aligned, multi-channel WAV — one transmitter per channel, paced through one real-time clock. For the canonical two-station setup that's a stereo file with one station's transmitted audio in each ear.

This is the feature requested: a WAV that is an accurate timeline of what actually happened on the channel — overlapping transmissions overlap in time, inter-burst gaps are real silence, and both channels share one start time. Drop it into Audacity and you can see at a glance who was keying when (ideal for hidden-node collisions — put the two hidden stations in left/right).

Why it's new, not the existing recorder

The per-port *.tx.wav files are written straight from each TNC as it bursts TX audio (samoyed can dump a 500 ms frame onto its UDP socket in a few wall-clock ms), so they carry the waveform faithfully but are not a real-time timeline. The composite recorder drains each transmitter's blocks through a single ticker at the block rate — the same mechanism rxFeeder already uses to pace audio into the receivers — writing one interleaved frame per tick. Every channel is advanced together by one goroutine, so they stay sample-aligned and share one start time; idle channels emit silence so gaps are preserved. The captured audio is the clean transmitter output (full-scale, before per-link loss / noise / the capture-effect mixer) — what each station put on the air, not what any one receiver heard.

Changes

• internal/audio — generalise the WAV header writer to N channels; add MultiWAVWriter and an Interleave helper.
• internal/router — compositeRecorder with per-channel queues fed from the txReader hot path (atomic pointer, mirroring the existing recorder) and a real-time pacer goroutine; Router.StartCompositeRecording / StopCompositeRecording / CompositeStatus; graceful drain + WAV-header patch on stop and on shutdown.
• cmd/sim-web — POST/GET /api/record/composite/{start,stop,status,download} for external software; a composite block in /api/status; a Left/Right-ear picker with Start/Stop/Download on the control page. Map UI untouched. Gated on -record DIR.
• cmd/sim-router — -composite a.vhf,b.vhf (requires -record) to capture on startup.
• README — documents the feature, CLI, and HTTP API.

API (for external software)

POST /api/record/composite/start with optional {"ports":["a.vhf","c.vhf"]} (omit → first two ports) → transmit over KISS → POST .../stop → GET .../download.

Testing

• Unit: multichannel WAV header + interleave round-trip; recorder start/stop/validation/defaults/real-time pacing; sim-web handlers via httptest.
• A concurrent-feed test passes under -race.
• End-to-end in the Docker image with live samoyed children: started a composite recording via the API, and a 0.7 s wall-clock capture produced exactly 0.70 s of valid stereo 44.1 kHz audio — confirming real-time pacing through the whole pipeline. (The PR's docker workflow rebuilds the image to validate it; merge to main publishes :main.)