[Test][TFLite] Add unit tests for LEAKY_RELU, HARD_SWISH ReLU_N1_to_1 and LOG_SOFTMAX

[fnhirwa]

This PR adds unit test coverage for activation function operators
in the Relax TFLite frontend, as part of #18971

• Added the (tf.nn.log_softmax, R.nn.log_softmax) entry to elementwise unit test
• Added unit test for LEAK_RELU with default alpha
• Added unit test for HARD_SWISH
• Added unit test for ReLU_N1_to_1

Bugs fixed:

• relax.op.tensor.clip -> relax.op.clip in _relu6 helper of convert_hard_swish
• relax.op.nn.leaky_relu -> relax.op.nn.leakyrelu in convert_leaky_relu

[gemini-code-assist]

Code Review

This pull request updates the TVM Relax TFLite frontend by renaming the leaky_relu operator to leakyrelu and updating the clip operator call. It also adds new test cases for leaky_relu, hard_swish, relu_n1_to_1, and log_softmax. The reviewer suggests using integer literals for the clip bounds in the hard_swish implementation and its associated test to ensure consistency with other activation functions and avoid potential overhead.

[gemini-code-assist]

For consistency with other activation functions implemented with clip in this file (like convert_relu6 and convert_relu_n1_to_1), it would be better to use integer literals for the clip bounds. This avoids potential overhead from using float64 bounds on a float32 tensor and improves consistency.

Suggested change

	return relax.op.clip(data, 0.0, 6.0)
	return relax.op.clip(data, 0, 6)

[gemini-code-assist]

To align with the suggested change of using integer clip bounds in convert_hard_swish, this expected IR should be updated. Using min and max arguments with integer values is also more consistent with test_relu_n1_to_1.

                lv1: R.Tensor((1, 30), dtype="float32") = R.clip(lv, min=0, max=6)

[tlopex]

LGTM! Thank you