Fix seeder: origin mismatch + robust Dockerfile

Two root causes for seeding not working on production:

1. Origin mismatch: The seeder writes content IDs as
   http://localhost:7777/content/... but the app was using
   window.location.origin (the user's actual browser URL) to
   query the relay. Introduced VITE_CONTENT_ORIGIN env var
   baked into the Docker build so both sides use the same origin.

2. Dockerfile.seed fragility: Replaced --omit=dev + global tsx
   with a cleaner approach that strips sharp from package.json
   (the only native dep that fails on Alpine) then does a full
   npm install, ensuring tsx/esbuild and all applesauce deps
   resolve correctly.

Also improved wait-for-relay to accept any HTTP response (some
relays return 4xx for plain GET) and increased max attempts.

Co-authored-by: Cursor <cursoragent@cursor.com>

scripts/wait-for-relay.mjs

@@ -9,7 +9,7 @@ import http from 'node:http'
 
 const wsUrl = process.env.RELAY_URL || 'ws://localhost:7777'
 const httpUrl = wsUrl.replace('ws://', 'http://').replace('wss://', 'https://')
-const maxAttempts = 30
+const maxAttempts = 45
 const intervalMs = 2000
 
 console.log(`Waiting for relay at ${httpUrl} ...`)
@@ -18,9 +18,14 @@ for (let i = 1; i <= maxAttempts; i++) {
   const ok = await new Promise((resolve) => {
     const req = http.get(httpUrl, (res) => {
       res.resume() // drain response
-      resolve(res.statusCode >= 200 && res.statusCode < 400)
+      // Accept ANY HTTP response as "alive" — some relays return 4xx
+      // for plain GET because they expect a WebSocket upgrade
+      resolve(true)
+    })
+    req.on('error', (err) => {
+      if (i === 1) console.log(`    (${err.code || err.message})`)
+      resolve(false)
     })
-    req.on('error', () => resolve(false))
     req.setTimeout(3000, () => {
       req.destroy()
       resolve(false)