[miopen-provider]: Add Integration tests to verify Batchnorm training rejects insufficient spatial

[bghimireamd]

Suggestion Description

Add integration tests in plugins/miopen_legacy_plugin/integration_tests/ to verify MIOpen plugin gracefully rejects batchnorm training when batch × spatial ≤ 1.

Test Cases:

1. Shape {1, C, 1, 1} (B×S = 1) → graph.build() returns error
2. Shape {0, C, H, W} (zero batch) → rejected

Acceptance Criteria:

• result.code != ErrorCode::OK
• No crashes or undefined behavior
• Error message mentions spatial dimension requirement
• Tests follow naming convention: IntegrationGpuBatchnormUnhappySpatial
• Tests added to plugins/miopen_legacy_plugin/integration_tests/

Note: This only applies to batchnorm training, not inference.

Reference: See getValidateSpatialDimensionsTestCases() in unit tests