http: skip byte length scan for latin1 writes

Single byte encodings have a byte length equal to the string length, so
measuring them with `Buffer.byteLength()` is wasted work. Every write
this module makes internally — the header block, the chunk size lines
and the trailer — is latin1, so they were all paying for a scan that
could only return `data.length`.

Check for those encodings before measuring. This takes the overhead of
the previous commit on a chunked write from ~13% to ~5%, and leaves
utf8 bodies, which do have to be measured, unaffected.

Refs: #57985
Signed-off-by: Ali Ahmed <ali.lah.aed456@gmail.com>

Author: QuantumBreakz

lib/_http_outgoing.js

@@ -465,8 +465,12 @@ function _writeRaw(data, encoding, callback, size) {
   // count of UTF-16 code units and undercounts any multi-byte character.
   // Callers that already computed the byte length hand it over as `size` so
   // that the string is not measured twice.
-  const len = size ?? (typeof data === 'string' ?
-    Buffer.byteLength(data, encoding) : data.length);
+  // Single byte encodings, which is what every internal write here uses,
+  // have a byte length equal to the string length; only utf8 has to be
+  // measured.
+  const len = size ?? (typeof data !== 'string' || encoding === 'latin1' ||
+    encoding === 'ascii' || encoding === 'binary' ?
+    data.length : Buffer.byteLength(data, encoding));
   this.outputSize += len;
   this._onPendingData(len);
   return this.outputSize < this[kHighWaterMark];