fix: set SO_REUSEADDR on port pre-check so restarts don't wait for TIME_WAIT

The fail-fast port check bound a plain socket, so TIME_WAIT sockets left
by a just-stopped instance (~60s) made it report 'port already in use'
even though uvicorn (which sets SO_REUSEADDR) could bind fine. Mirror
uvicorn's bind semantics: the check now only fails when the port is
genuinely held by a live listener.

nvcurve/server.py

@@ -2610,7 +2610,11 @@ def run(
 
     # Fail fast if the port is already in use — silently shifting ports breaks
     # client discovery. Users should configure a different port explicitly.
+    # SO_REUSEADDR mirrors uvicorn's own bind: without it, TIME_WAIT sockets
+    # left by a just-stopped instance (~60s) would fail this check even though
+    # uvicorn could bind fine, forcing a wait on every restart.
     with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+        s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
         try:
             s.bind((host, port))
         except OSError: