Import march-2020-03-10 from private repo (scrubbed 2026-05-27)

Targets Rec Room 2020.03.10 client (also serves 2020.03.06).

Re-import after adding:
- DorkNet.ClientMod/dorknet-clientmod.json.template — config schema
  the launcher's manifest-driven patcher renders into UserData/.
- tools/build-patcher-zip.ps1 — produces dorknet-clientpatch-{key}.zip
  for v1-march-* Release uploads.
- .github/workflows/release.yml — on tag push publishes both
  dorknet-server-march_2020_03_10-win-x64.zip and
  dorknet-clientpatch-march_2020_03_10.zip as Release assets.

Same scrub rules as december; tools/build-patcher-zip.ps1 added to
the maintainer-side allow-list so it survives.

DorkNet.Server/Controllers/Health/HealthController.cs

@@ -0,0 +1,66 @@
+using Microsoft.AspNetCore.Mvc;
+using Microsoft.EntityFrameworkCore;
+using DorkNet.Server.Data;
+
+namespace DorkNet.Server.Controllers.Health;
+
+/// <summary>
+/// GET <c>/healthz</c> — liveness/readiness probe for Coolify rolling deploys.
+///
+/// Returns 503 until BOTH conditions hold:
+/// 1. <see cref="MigrationsComplete"/> flag has been set by Program.cs after
+///    <c>db.Database.Migrate()</c> returns (so a new replica that's still
+///    running migrations doesn't get traffic shifted to it before its schema
+///    is current).
+/// 2. <c>db.Database.CanConnectAsync()</c> succeeds (catches transient DB
+///    blips and prevents traffic during a Postgres restart).
+///
+/// Coolify's rolling-deploy strategy waits for the new container's healthz
+/// to return 200 before draining the old one. Combined with a brief
+/// (~5–10s) `start-first` overlap, this is the closest thing to zero
+/// downtime we get with single-region hosting.
+///
+/// HostFilteringOptions is configured (via DomainConfig) to accept
+/// <c>localhost</c> + <c>127.0.0.1</c> alongside the apex pool so the
+/// Coolify internal healthcheck (which hits <c>localhost:8080/healthz</c>
+/// from inside the container) reaches this handler — earlier
+/// per-controller <c>[Host(...)]</c> filters would have 404ed it.
+///
+/// Distinct from <c>/v1/ping</c> on the ns subdomain — that returns a
+/// timestamped payload the watch's keep-alive loop reads and is part of
+/// the wire contract. We deliberately don't change it.
+/// </summary>
+[ApiController]
+public class HealthController(DorkNetDbContext db) : ControllerBase
+{
+    /// <summary>Set to true by Program.cs after the EF Core
+    /// <c>Migrate()</c> call returns successfully. Volatile because
+    /// it's read from request threads concurrently with the startup
+    /// thread's write. Process-local — every replica sets its own
+    /// flag once its own startup migrations finish.</summary>
+    public static volatile bool MigrationsComplete = false;
+
+    // Order is more-negative than CdnController.Serve's wildcard
+    // (Order = -100). Without this the catch-all matched <c>/healthz</c>
+    // first on the <c>localhost</c> host (no subdomain) and returned
+    // 404 → Docker's HEALTHCHECK <c>curl --fail</c> exited non-zero →
+    // Coolify rolled the new container back.
+    [HttpGet("/healthz", Order = -1000)]
+    public async Task<IActionResult> Healthz()
+    {
+        if (!MigrationsComplete)
+            return StatusCode(503, new { status = "starting", reason = "migrations_pending" });
+
+        // CanConnectAsync uses a cheap "SELECT 1"-equivalent — fast enough
+        // to run on every probe (~1ms for a warm pool). Cheaper than the
+        // alternative of failing only on actual query timeouts.
+        bool dbOk;
+        try { dbOk = await db.Database.CanConnectAsync(); }
+        catch { dbOk = false; }
+
+        if (!dbOk)
+            return StatusCode(503, new { status = "unhealthy", reason = "db_unreachable" });
+
+        return Ok(new { status = "ok" });
+    }
+}