perf(benchmark): optimize nginx and caddy configs for raw throughput

nginx: add aio threads, open_file_cache, directio, sendfile_max_chunk,
output_buffers, bumped keepalive_requests to 10k, listen backlog 4096,
server_tokens off, reset_timedout_connection.

caddy: disable admin API, explicit auto_https off, suppress Server and
X-Content-Type-Options headers, disable_canonical_uris in file_server.

Author: jkyberneees

benchmark/Caddyfile

@@ -1,8 +1,27 @@
+{
+    # Global options
+    # Disable the admin API — not needed for benchmarking
+    admin off
+
+    # Prevent Caddy from trying HTTPS / ACME
+    auto_https off
+}
+
 :80 {
     root * /www
-    file_server
+    file_server {
+        # Pre-compress is off — fair raw-throughput comparison
+        disable_canonical_uris
+    }
+
+    # Compression off — fair comparison with raw file serving
+    # (encode gzip / encode zstd deliberately omitted)
+
+    # Suppress verbose response headers
+    header -Server
+    header -X-Content-Type-Options
 
-    # Disable access log for fair benchmark comparison
+    # Disable access log
     log {
         output discard
     }

benchmark/nginx.conf

@@ -1,29 +1,48 @@
-worker_processes auto;
-worker_rlimit_nofile 65535;
+worker_processes      auto;
+worker_rlimit_nofile  65535;
 
 events {
-    worker_connections 4096;
-    use epoll;
-    multi_accept on;
+    worker_connections  4096;
+    use                 epoll;
+    multi_accept        on;
 }
 
 http {
     include       /etc/nginx/mime.types;
     default_type  application/octet-stream;
 
-    sendfile           on;
-    tcp_nopush         on;
-    tcp_nodelay        on;
-    keepalive_timeout  65;
-    keepalive_requests 1000;
+    # ── I/O ──────────────────────────────────────────────────────────────────
+    sendfile            on;
+    sendfile_max_chunk  512k;   # cap per-call to avoid worker starvation
+    aio                 threads; # async file I/O via thread pool (built-in)
+    directio            4m;     # bypass page cache for files > 4 MB
+    tcp_nopush          on;     # batch headers + first data chunk (needs sendfile)
+    tcp_nodelay         on;     # flush remaining small frames immediately
 
-    access_log off;
-    error_log  /dev/null crit;
+    # ── Keepalive ────────────────────────────────────────────────────────────
+    keepalive_timeout   75;
+    keepalive_requests  10000;
+    reset_timedout_connection on;
 
-    gzip off;  # disabled — fair comparison with raw file serving
+    # ── Buffers ──────────────────────────────────────────────────────────────
+    output_buffers      2 128k;
+
+    # ── Open-file cache (avoid stat/open syscalls on every request) ──────────
+    open_file_cache          max=10000 inactive=30s;
+    open_file_cache_valid    60s;
+    open_file_cache_min_uses 1;
+    open_file_cache_errors   on;
+
+    # ── Logging / noise ──────────────────────────────────────────────────────
+    access_log   off;
+    error_log    /dev/null crit;
+    server_tokens off;
+
+    # ── Compression — disabled for fair raw-throughput comparison ────────────
+    gzip off;
 
     server {
-        listen 80;
+        listen 80 backlog=4096;
         root   /www;
         index  index.html;