webnn: add wpt for bidirectional lstm

Fixes some minor bugs with CoreML bidrection implementation.

Also adjusted ULP for this complex op to 3, this is same as gru.

Bug: 360052663
Change-Id: Ie15ab15a359be70e219fd61ee0d3a3fea21f3405
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/6103076
Commit-Queue: Phillis Tang <phillis@chromium.org>
Reviewed-by: Reilly Grant <reillyg@chromium.org>
Cr-Commit-Position: refs/heads/main@{#1399398}

webnn/conformance_tests/lstm.https.any.js

@@ -50,7 +50,7 @@
 
 
 const getLstmPrecisionTolerance = (graphResources) => {
-  const toleranceValueDict = {float32: 1, float16: 1};
+  const toleranceValueDict = {float32: 3, float16: 3};
   const expectedDataType =
       graphResources
           .expectedOutputs[Object.keys(graphResources.expectedOutputs)[0]]
@@ -866,6 +866,90 @@ const lstmTests = [
       }
     }
   },
+  {
+    'name': 'lstm float32 tensors steps=2 with bidirections',
+    'graph': {
+      'inputs': {
+        'lstmInput': {
+          'data': [1, 2, 2, 1, 3, 4, 1, 2],
+          'descriptor': {shape: [2, 2, 2], dataType: 'float32'}
+        },
+        'lstmWeight': {
+          'data': [
+            1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2,
+            1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2
+          ],
+          'descriptor': {shape: [2, 8, 2], dataType: 'float32'},
+          'constant': true
+        },
+        'lstmRecurrentWeight': {
+          'data': [
+            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
+            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
+            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1
+          ],
+          'descriptor': {shape: [2, 8, 2], dataType: 'float32'},
+          'constant': true
+        },
+        'lstmBias': {
+          'data': [1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2],
+          'descriptor': {shape: [2, 8], dataType: 'float32'},
+          'constant': true
+        },
+        'lstmRecurrentBias': {
+          'data': [1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2],
+          'descriptor': {shape: [2, 8], dataType: 'float32'},
+          'constant': true
+        }
+      },
+      'operators': [{
+        'name': 'lstm',
+        'arguments': [
+          {'input': 'lstmInput'}, {'weight': 'lstmWeight'},
+          {'recurrentWeight': 'lstmRecurrentWeight'}, {'steps': 2},
+          {'hiddenSize': 2}, {
+            'options': {
+              'bias': 'lstmBias',
+              'recurrentBias': 'lstmRecurrentBias',
+              'returnSequence': true,
+              'direction': 'both',
+              'layout': 'iofg'
+            }
+          }
+        ],
+        'outputs': ['lstmOutput1', 'lstmOutput2', 'lstmOutput3']
+      }],
+      'expectedOutputs': {
+        'lstmOutput1': {
+          'data': [
+            0.5764073133468628, 0.8236227035522461, 0.6612355709075928,
+            0.8442635536193848, 0.5764073133468628, 0.8236227035522461,
+            0.8635294437408447, 0.9491351246833801
+          ],
+          'descriptor': {shape: [2, 2, 2], dataType: 'float32'}
+        },
+        'lstmOutput2': {
+          'data': [
+            1.0171456336975098, 1.6205494403839111, 1.3388464450836182,
+            1.7642604112625122, 1.0171456336975098, 1.6205494403839111,
+            1.4856269359588623, 1.8449554443359375
+          ],
+          'descriptor': {shape: [2, 2, 2], dataType: 'float32'}
+        },
+        'lstmOutput3': {
+          'data': [
+            0.3696063756942749, 0.6082833409309387, 0.7037754058837891,
+            0.7586681246757507, 0.5764073133468628, 0.8236227035522461,
+            0.8635294437408447, 0.9491351246833801, 0.5764073133468628,
+            0.8236227035522461, 0.6612355709075928, 0.8442635536193848,
+            0.3696063756942749, 0.6082833409309387, 0.3696063756942749,
+            0.6082833409309387
+          ],
+          'descriptor': {shape: [2, 2, 2, 2], dataType: 'float32'}
+        }
+      }
+    }
+  },
   {
     'name':
         'lstm float16 tensors steps=1 with options.bias, options.recurrentBias',