fix(material/table): improve filter predicate efficiency (#30172)

(cherry picked from commit c12b73e)

src/material/table/table-data-source.ts

@@ -229,23 +229,12 @@ export class MatTableDataSource<T, P extends MatPaginator = MatPaginator> extend
    * @returns Whether the filter matches against the data
    */
   filterPredicate: (data: T, filter: string) => boolean = (data: T, filter: string): boolean => {
-    // Transform the data into a lowercase string of all property values.
-    const dataStr = Object.keys(data as unknown as Record<string, any>)
-      .reduce((currentTerm: string, key: string) => {
-        // Use an obscure Unicode character to delimit the words in the concatenated string.
-        // This avoids matches where the values of two columns combined will match the user's query
-        // (e.g. `Flute` and `Stop` will match `Test`). The character is intended to be something
-        // that has a very low chance of being typed in by somebody in a text field. This one in
-        // particular is "White up-pointing triangle with dot" from
-        // https://en.wikipedia.org/wiki/List_of_Unicode_characters
-        return currentTerm + (data as unknown as Record<string, any>)[key] + '◬';
-      }, '')
-      .toLowerCase();
-
     // Transform the filter by converting it to lowercase and removing whitespace.
     const transformedFilter = filter.trim().toLowerCase();
-
-    return dataStr.indexOf(transformedFilter) != -1;
+    // Loops over the values in the array and returns true if any of them match the filter string
+    return Object.values(data as {[key: string]: any}).some(value =>
+      `${value}`.toLowerCase().includes(transformedFilter),
+    );
   };
 
   constructor(initialData: T[] = []) {