Replace buffer.slice in text search with buffer.toString(undefined, start, end).

This is much faster although it's not totally obvious why. But it's run for every line of every file in the workspace so it's very sensitive.  See #15384.

src/vs/workbench/services/search/node/textSearch.ts

@@ -166,16 +166,16 @@ export class Engine implements ISearchEngine<ISerializedFileMatch> {
 
 			const outer = this;
 
-			function decodeBuffer(buffer: NodeBuffer): string {
+			function decodeBuffer(buffer: NodeBuffer, start: number, end: number): string {
 				if (options.encoding === UTF8 || options.encoding === UTF8_with_bom) {
-					return buffer.toString(); // much faster to use built in toString() when encoding is default
+					return buffer.toString(undefined, start, end); // much faster to use built in toString() when encoding is default
 				}
 
-				return decode(buffer, options.encoding);
+				return decode(buffer.slice(start, end), options.encoding);
 			}
 
 			function lineFinished(offset: number): void {
-				line += decodeBuffer(buffer.slice(pos, i + offset));
+				line += decodeBuffer(buffer, pos, i + offset);
 				perLineCallback(line, lineNumber);
 				line = '';
 				lineNumber++;
@@ -245,7 +245,7 @@ export class Engine implements ISearchEngine<ISerializedFileMatch> {
 						}
 					}
 
-					line += decodeBuffer(buffer.slice(pos, bytesRead));
+					line += decodeBuffer(buffer, pos, bytesRead);
 
 					readFile(false /* isFirstRead */, clb); // Continue reading
 				});