[ORT CPU][WebNN] Fail to run some WebNN batchNormalization tests due to tolerance issues

[BruceDai]

Describe the issue

Fail to run below WebNN batchNormalization tests due to tolerance issues by ORT default CPU EP.

Test	Error message
batchNormalization float32 2D tensor (mean and variance are non-constant) default options	assert_less_than_equal: assert_array_approx_equals_ulp: test batchNormalization float32 actual 1.8526287078857422 should be close enough to expected 1.8526301383972168 by ULP distance: expected a number less than or equal to 6n but got 12n
batchNormalization float32 2D constant tensors default options	assert_less_than_equal: assert_array_approx_equals_ulp: test batchNormalization float32 actual 1.8526287078857422 should be close enough to expected 1.8526301383972168 by ULP distance: expected a number less than or equal to 6n but got 12n
batchNormalization float32 2D tensor default options	assert_less_than_equal: assert_array_approx_equals_ulp: test batchNormalization float32 actual 1.8526287078857422 should be close enough to expected 1.8526301383972168 by ULP distance: expected a number less than or equal to 6n but got 12n
[optional] batchNormalization default + relu	assert_less_than_equal: assert_array_approx_equals_ulp: test batchNormalization relu float32 actual 1.8526287078857422 should be close enough to expected 1.8526301383972168 by ULP distance: expected a number less than or equal to 6n but got 12n
[optional] batchNormalization default + leakyRelu	assert_less_than_equal: assert_array_approx_equals_ulp: test batchNormalization leakyRelu float32 actual 1.8526287078857422 should be close enough to expected 1.8526301383972168 by ULP distance: expected a number less than or equal to 7n but got 12n
[optional] batchNormalization default + clamp	assert_less_than_equal: assert_array_approx_equals_ulp: test batchNormalization clamp float32 actual 1.8526287078857422 should be close enough to expected 1.8526301383972168 by ULP distance: expected a number less than or equal to 6n but got 12n
[optional] batchNormalization default + prelu	assert_less_than_equal: assert_array_approx_equals_ulp: test batchNormalization prelu float32 actual 1.8526287078857422 should be close enough to expected 1.8526301383972168 by ULP distance: expected a number less than or equal to 7n but got 12n
[required] batchNormalization options.axis=0 + gelu	assert_less_than_equal: assert_array_approx_equals_ulp: test batchNormalization gelu float32 actual -0.035951320081949234 should be close enough to expected -0.03595117852091789 by ULP distance: expected a number less than or equal to 24n but got 38n
[optional] batchNormalization default + hardSwish	assert_less_than_equal: assert_array_approx_equals_ulp: test batchNormalization hardSwish float32 actual 1.4983532428741455 should be close enough to expected 1.4983549118041992 by ULP distance: expected a number less than or equal to 10n but got 14n
[optional] batchNormalization default + linear	assert_less_than_equal: assert_array_approx_equals_ulp: test batchNormalization linear float32 actual 1.8526287078857422 should be close enough to expected 1.8526301383972168 by ULP distance: expected a number less than or equal to 8n but got 12n

/cc @fdwr PTAL, thanks!

To reproduce

1. Install Windows App SDK Stable 1.8.2 (1.8.251003001)
2. Install lates Edge Canary browser
3. Launch Edge Canary browser, navigate to about://flags enabled "Enables WebNN API" and "ONNX Runtime backend for WebNN" flags, then relaunch browser
4. Run each failure test using below test url

Test	Test URL
batchNormalization float32 2D tensor (mean and variance are non-constant) default options	https://wpt.live/webnn/conformance_tests/batch_normalization.https.any.html?device=cpu&tc=batchNormalization%20float32%202D%20tensor%20(mean%20and%20variance%20are%20non-constant)%20default%20options
batchNormalization float32 2D constant tensors default options	https://wpt.live/webnn/conformance_tests/batch_normalization.https.any.html?device=cpu&tc=batchNormalization%20float32%202D%20constant%20tensors%20default%20options
batchNormalization float32 2D tensor default options	https://wpt.live/webnn/conformance_tests/batch_normalization.https.any.html?device=cpu&tc=batchNormalization%20float32%202D%20tensor%20default%20options
batchNormalization default + relu	https://wpt.live/webnn/conformance_tests/subgraph.https.any.html?device=cpu&tc=batchNormalization%20default%20%2B%20relu
batchNormalization default + leakyRelu	https://wpt.live/webnn/conformance_tests/subgraph.https.any.html?device=cpu&tc=batchNormalization%20default%20%2B%20leakyRelu
batchNormalization default + clamp	https://wpt.live/webnn/conformance_tests/subgraph.https.any.html?device=cpu&tc=batchNormalization%20default%20%2B%20clamp
batchNormalization default + prelu	https://wpt.live/webnn/conformance_tests/subgraph.https.any.html?device=cpu&tc=batchNormalization%20default%20%2B%20prelu
batchNormalization options.axis=0 + gelu	https://wpt.live/webnn/conformance_tests/subgraph.https.any.html?device=cpu&tc=batchNormalization%20options.axis=0%20%2B%20gelu
batchNormalization default + hardSwish	https://wpt.live/webnn/conformance_tests/subgraph.https.any.html?device=cpu&tc=batchNormalization%20default%20%2B%20hardSwish
batchNormalization default + linear	https://wpt.live/webnn/conformance_tests/subgraph.https.any.html?device=cpu&tc=batchNormalization%20default%20%2B%20linear

Urgency

No response

ONNX Runtime Installation

Released Package Windows App SDK Stable 1.8.2 (1.8.251003001)

ONNX Runtime Version or Commit ID

1.23.25.928

Execution Provider

ORT default CPU EP