infra: Railway Vivado self-hosted GH Actions runner

[gHashTag]

Goal

Provision a managed x86_64 Linux runner with Vivado 2025.2 installed, registered as a self-hosted GitHub Actions runner against this repo. Unblocks PR #604 (Vivado-CI with UART telemetry) and tok/s measurement on silicon.

Constraints (verified)

• M4 MacBook Pro (228 GB SSD, ~11 GB free, ARM64) cannot host Vivado: needs ~110 GB and x86_64 Linux.
• Vivado_2025.2_Lin64.bin (347 MB installer) already in ~/Downloads on Mac.
• Project policy: no .sh / .py anywhere — Rust only (inline awk in YAML is OK).

Solution shipped in this branch

infra/railway-vivado-runner/:

• Dockerfile — Ubuntu 22.04 + Vivado deps + GH Actions runner v2.319.1
• src/main.rs — Rust entrypoint that registers the runner and exec's run.sh (project-supplied GH binary)
• Cargo.toml — entrypoint crate
• railway.toml — service definition with 3 persistent volumes (130 GB Xilinx, 1 GB installer, 20 GB work)
• README.md — step-by-step Railway deploy (volumes, env vars, Vivado silent install)

.github/workflows/vivado-self-hosted.yml:

• runs-on: [self-hosted, vivado, x86_64, linux]
• Builds t27c, generates Verilog from .t27 specs, runs Vivado synth, optionally runs UART smoke if DLC-10 attached
• Triggers on PR + manual dispatch with design selection (blinky / gf16 / phi_heartbeat) and UART toggle

Cost estimate

Railway Pro, 8 vCPU / 32 GB / 150 GB volume: ~$35–55/month active.

Next actions for operator

1. railway login && railway init from infra/railway-vivado-runner/
2. Provision 3 volumes per README
3. railway volume push ~/Downloads/Vivado_2025.2_Lin64.bin /opt/installer/
4. Get fresh runner registration token from repo Settings → Actions → Runners
5. Set 4 env vars, railway up
6. One-time Vivado silent install via railway run
7. Verify runner shows Idle / Online in repo settings

Verification path

• Trigger Vivado workflow with design=blinky, expect bitstream + utilization summary in artifacts
• If DLC-10 attached, expect tok_s.txt with measured tokens/sec

Refs #604

[gHashTag]

Closed — replaced by GHA Docker approach in #622. No Railway provisioning needed; Vivado runs as a pre-built ghcr.io image on ubuntu-latest.