perf(tools): stop counting string code points once the bounds are decided

Strings without length bounds are not counted, and counting stops at the
larger bound, so a huge tool argument costs no more than its limit.

src/tools/tool-schema-codepoints.test.mjs

@@ -18,3 +18,16 @@ test('tool string schemas count Unicode characters rather than UTF-16 units', ()
   );
   assert.equal(validateValue({ type: 'string', minLength: 2 }, '𠮷').length, 1);
 });
+
+test('tool string bounds hold at their edges and on huge arguments', () => {
+  const string = (bounds) => ({ type: 'string', ...bounds });
+  assert.deepEqual(validateValue(string({ maxLength: 3 }), '😀😀😀'), []);
+  assert.equal(validateValue(string({ maxLength: 3 }), '😀😀😀😀').length, 1);
+  assert.deepEqual(validateValue(string({ minLength: 3 }), '😀😀😀'), []);
+  assert.deepEqual(validateValue(string({}), 'x'.repeat(1e6)), []);
+  // Counting all of it takes roughly 0.7 s; stopping at the limit takes none.
+  const huge = 'x'.repeat(1e8);
+  const started = performance.now();
+  assert.equal(validateValue(string({ maxLength: 60 }), huge).length, 1);
+  assert.ok(performance.now() - started < 200, 'stops counting past the limit');
+});